Professional Documents
Culture Documents
08自由意志 Peter W. Kalivas, Martin P. Paulus - Intrusive Thinking - From Molecules to Free Will-MIT Press (2021)
08自由意志 Peter W. Kalivas, Martin P. Paulus - Intrusive Thinking - From Molecules to Free Will-MIT Press (2021)
Got Mad at Me
Have you ever wondered why thoughts appear and can motivate you to do
things you wouldn’t otherwise do? One afternoon, Sue (my wife) called me at
work to tell me she was going out with friends after work, and it was up to me
to get dinner and do homework with the kids. Sue and her messages were (and
remain to this day) very motivating for me, so I packed up and went down to
my car. On the way to the car and during the drive home, I randomly thought
about what I would cook for dinner, which kid was having a test and would
need help, and various other plans for the evening. About halfway home,
Kyle from the pub called. He was energized because a buddy from southern
California had just flown in and thus Kyle was rounding people up for what
sounded like a really fun night. I was going to tell him, “Sorry, I can’t make it,”
but then the 20 beers on tap suddenly popped into my mind, as did the hilari-
ous time we had the last time we all got together. Pretty soon I wasn’t thinking
about dinner plans or school tests any more. I was making plans to be with my
friends and thinking about the latest release IPA from the local microbrewery.
When we ended our call, I had told him that I wasn’t sure what I would do.
At this point, we might ask ourselves what is more motivating: going home to
cooking and homework, or going to the pub for beer and friends. Each one of
us faces these types of decisions, big and small, all the time, and what we do
depends largely on how we organize our thoughts about each possible scenario
and which ultimately seems most important to us. In this story, if you know
Sue, you pretty much know already that I went home and had to hear about
all of the pub hilarity the next morning. But, if I suffered from alcohol use
disorder, thoughts about what would happen at the pub, the taste of the beer,
and past really fun pub experiences would have inevitably intruded until they
all but squeezed out thoughts of dinner and homework with my kids. My plans
to go home would begin to fade until they were all but forgotten, or perhaps
I would rationalize that stopping by for one of the special IPAs before going
home would somehow work out.
The story above illustrates how thoughts of pub friends and beer can intrude
in substance use disorder. However, the intrusion of traumatic events in post-
traumatic stress disorder, rumination on negative outcomes in depression, or
hearing voices in schizophrenia are all examples of thoughts generated by your
brain that can contribute to debilitating psychiatric disorders. Of course, it is a
natural and healthy adaptive process to produce thoughts either randomly or in
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
xvi P. W. Kalivas
association with the world we are experiencing, and then to use these thoughts
to navigate successfully toward desired outcomes.
This volume explores and provides the best possible explanations for what
this process is, how it gets usurped in psychiatric disorders, and what this
knowledge of how the brain handles thoughts means for concepts of free will
and one’s responsibility for poor decisions, especially when a thought disorder
exists. It addresses how the brain is organized to create thoughts that can be ig-
nored or can build in motivational content, and how we then weigh thoughts to
decide on behavior that best adapts us to the world. It also poses and attempts
to answer a number of questions that are commonly asked: How do the mecha-
nisms of thought intrusion and decision making get corrupted in psychiatric
disorders to create intrusions that cannot be controlled? How are thought intru-
sions usurped by motivation to produce behavior that may be maladaptive, at
least according to social norms? What is free will and what responsibility does
free will (or lack of it) create for how we behave?
— Peter W. Kalivas
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Molecules and Circuits
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
2
Abstract
This chapter discusses psychological constructs considered to be central to the media-
tion of intrusive thinking and the neural circuits that underlie these processes. It assimi-
lates intrusive thoughts with conditioned responses, discerns associate structures that
can support these responses, and suggests how episodic information may be integrated
with these associations. Mechanisms by which intrusive thoughts can be updated are
explored, with a focus on extinction and memory reconsolidation. Intrusive thoughts
ultimately engage many areas of the brain as they encompass sensory, cognitive, mo-
tor, and somatic processes. In this chapter, the focus is on specific circuits within the
prefrontal-limbic network that are proposed to encode, update, and maintain the con-
tent of intrusions. These circuits include interconnecting pathways between the ventral
tegmental area, nucleus accumbens, medial prefrontal and orbitofrontal cortices, hip-
pocampus, and the amygdaloid complex.
Introduction
To begin to answer the question posed in the title, we must first consider what
intrusive thinking is. Specifically, we need to address the nature of the con-
tent of an intrusive thought: Is it an episodic representation? Could it be a
visceral urge, an emotive state that is devoid of episodic content? There is
unlikely to be a single answer, as intrusive thinking is ultimately made up of
different compositions of these extremes, specific to the underlying pathology.
For example, delusions in schizophrenia are clearly episodic in nature, often
composed of a detailed and elaborate narrative. By contrast, in obsessive-com-
pulsive disorder (OCD), obsessions come more in the form of an urgency state
that is often likened to anxiety. While there may be episodic aspects to this
process, such as obsessing over specific contexts, the debilitating qualities of
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
12 P. E. M. Phillips and A. L. Milton
the obsession emerge from the anxiety-like state. Likewise, obsessive behavior
in substance use disorders and other addictive disorders comes in the form of
craving, an emotive state which, in its fundamental form, is devoid of episodic
content. This state, however, is intimately attached to episodic representation
of experiences related to the addictive behavior. Further, posttraumatic stress
disorder (PTSD) is also associated with an anxiety-like state that is very clearly
linked to fragmented episodic memories. This differentiation between the con-
tent of intrusive thoughts may, on the surface, seem subtle but is likely to be
critical when its neural substrates are considered because the circuits that pro-
cess episodic information are separable from those underlying emotive states.
These differences fall into a common dichotomy in the control of behavior;
namely, there are parallel systems that subserve cognitive functions. This di-
chotomy has a classic separation of processes that can be loosely summarized
in the form of a speed–accuracy trade-off: fast, low-computational processes
that amount to estimations are generated in parallel to more precise computa-
tions that have a higher cognitive demand. We and others have equated this
dichotomy to parallel processes used in machine learning that are classified as
model-free and model-based computations (Clark et al. 2012) as we believe
this is an intuitive and tractable framework. The basis of this separation is the
complexity of the information that is stored to support a learned association.
A model-based computation establishes an environmental model that can be
used to explore potential and inferred connections between stimuli and states.
In contrast, a model-free computation is a single-dimensional value assigned
to a stimulus based on the reliability of its association with a motivationally
relevant outcome.
Central to this line of reasoning, intrusive thoughts are often triggered by
environmental stimuli. For instance, in substance use disorders, drug craving
can be elicited by drug cues and is posited to become stimulus bound during the
transition to addiction (Tiffany and Carter 1998). Accordingly, drug seeking in
rodent models, a proxy for craving, can be elicited by unconditioned stimuli
(e.g., abused substances or stressors) or conditioned stimuli (e.g., drug cues or
conditioned stressors). In PTSD, intrusive episodes are often linked to envi-
ronmental stimuli in a manner consistent with overgeneralization (i.e., when
otherwise neutral stimuli elicit a threat response). Hence, intrusive thoughts
often take the form of Pavlovian-conditioned responses, which is our focus in
this chapter. As an aside, it is worth noting that while intrusive thoughts can
be triggered by unconditioned stimuli, they are unlikely to be unconditioned
responses because they are not elicited in naïve individuals but rather develop
with psychiatric pathology. Importantly, where tested, stimulus-driven intru-
sions exhibit similar neural signatures to those that are not triggered by an
explicit external stimulus (M. C. Anderson, pers. comm.).
What neural circuits are necessary to support these associations? We will
discuss circuits that support Pavlovian associations, both for emotive re-
sponses and those that incorporate representation of stimulus properties. We
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Circuits That Mediate and Update Intrusive Thinking 13
will also consider structures that mediate the storage and retrieval of episodic
memories as well as the interactions between all of these circuits. We will
make the case that circuitry, including midbrain dopamine neurons that project
to the nucleus accumbens in the ventral striatum (mesolimbic pathway) or to
the medial prefrontal cortex (mesocortical pathway), is necessary for at least
a subset of emotive associations. Thereafter, discussion of the neural circuitry
will be broadened to include substrates that support Pavlovian associations
that can support inferential reasoning, with a specific focus on the central role
of the orbitofrontal cortex (OFC) in these higher cognitive processes. We will
also explore medial temporal and frontal structures implicated in the acqui-
sition and consolidation of episodic memories and discuss circuits involved
in the process of updating existing associations, both through extinction and
memory reconsolidation.
Many psychiatric disorders that are associated with intrusive thinking (e.g.,
schizophrenia, substance use, OCD, and PTSD) have been linked to perturba-
tions in dopamine transmission in the striatum and/or the prefrontal cortex.
These clues have driven extensive research on dopamine transmission with
regard to psychiatric disorders, including those where intrusive thinking is a
prominent feature.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
14 P. E. M. Phillips and A. L. Milton
Aversive Signaling
To date, the focus has been on associations with appetitive stimuli. Needless
to say, intrusive thoughts are often evoked by aversive stimuli. The role of
the mesolimbic and mesocortical dopamine systems in computations relating
to aversive information has been much more controversial. In many cases,
it is simply assumed that aversive stimuli will be computed in this learn-
ing model in a similar manner to stimuli that predict lower than previously
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Circuits That Mediate and Update Intrusive Thinking 15
expected rewards. However, the evidence for this type of encoding is mixed.
Mirenowicz and Schultz (1996) observed minimal responses to mildly aversive
stimuli (air puffs to the hand) in midbrain dopamine neurons of nonhuman pri-
mates. In contrast, Matsumoto and Hikosaka (2009) observed robust changes
in the activity of dopamine cells on the presentation of aversive stimuli. In
this latter work, the investigators reported some populations of dopamine neu-
rons that encoded aversive information by changing their firing rates in the
opposite direction to reward information. Specifically, the presentation of un-
expected aversive stimuli or conditioned stimuli that increased the expectation
of aversion resulted in reduced firing of these neurons. However, Matsumoto
and Hikosaka also reported populations of putative dopamine neurons that in-
creased their firing rate to predictions of aversion. These neurons tended to
reside in dorsolateral aspects of the dopaminergic ventral midbrain (dorsolat-
eral substantia nigra pars compacta). They responded similarly to stimuli that
increased the expectation of aversion to those that increased the expectation of
reward, an encoding pattern often referred to as an unsigned prediction error.
However, this coding pattern is not without controversy. Specifically, Fiorillo
(2013) has argued that the observed positive responses to aversive stimuli re-
late to their sensory properties rather than their motivational salience. Others
have also questioned whether all of the recorded neurons in these studies are
truly dopamine-containing neurons. To address this concern of neuronal-type
specificity, Cohen et al. (2012) recorded the firing rates of genetically identi-
fied dopamine neurons in mice in response to presentations of appetitive- and
aversive-related stimuli. They reported that modulation of the firing rates to
aversive-related stimuli were consistently in the opposite direction to those
for reward stimuli. These neurons, however, were exclusively recorded in
the ventral tegmental area (i.e., the ventral medial aspect of the ventral mid-
brain) and so did not include the homologous anatomical region from which
the unsigned prediction error signals were reported by the Hikosaka group.
In neurochemical studies that measure dopamine levels in terminals, results
with aversive stimuli have also been mixed. Studies measuring dopamine lev-
els over minutes tend to report increases in dopamine to the presentation of
aversive stimuli, especially in the prefrontal cortex (Young 2004; Butts et al.
2011). In contrast, measurements on the order of seconds reveal decreases in
dopamine in the nucleus accumbens to aversive stimuli (Roitman et al. 2008).
Some investigators addressing this issue have argued that increases in dopa-
mine transmission following an aversive stimulus observed over minutes were
responses to the relief from aversion at the offset of the stimulus rather than a
response to the onset (Ungless 2004).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
16 P. E. M. Phillips and A. L. Milton
Broader Circuity
When considering the entirety of the pathways which support these stimu-
lus–stimulus associations, the complexity of the circuitry rapidly expands. In
addition to the laterodorsal tegmentum and the lateral habenula, the central
nucleus of the amygdala, in particular, has been implicated as important up-
stream circuitry to dopamine neurons in model-free processes (Clark et al.
2012). Nonetheless, it is important not to dismiss the rich convergent inputs
into the ventral midbrain from many areas of the brain (Geisler and Wise
2008). Indeed, an optimal temporal difference algorithm should have access to
all available sources of predictive information about rewards and punishers as
well as information on motivational states.
Orbitofrontal Cortex
From a plethora of research, it is evident that the OFC encodes many differ-
ent features of motivational stimuli (Thorpe et al. 1983; Padoa-Schioppa and
Assad 2006; Stalnaker et al. 2014). Indeed, it seems that just about any as-
pect of perception is encoded in about twenty percent of OFC neurons! This
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Circuits That Mediate and Update Intrusive Thinking 17
Episodic Information
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
18 P. E. M. Phillips and A. L. Milton
Extinction
To think about the updating of intrusive thoughts, we should consider the dif-
ferent plasticity processes that can be engaged following the consolidation of
an intrusive thought or memory. Broadly speaking, the neural trace or “en-
semble” that encodes a thought or memory can be triggered to induce retrieval,
reconsolidation, or extinction of the trace under similar, but importantly differ-
ent, conditions. Extinction has been the most extensively studied, having been
originally described by Pavlov (1927). Extinction is operationally defined as
the degradation of behavior that was previously supported by a learned asso-
ciation. It takes place when the reliability of the association is weakened, typi-
cally occurring after extensive exposure to the unreinforced cue. Extinction
can be modeled as a reduction in the strength of an association between a con-
dition and unconditioned stimulus in a simple bidirectional learning system,
such as the model-free system putatively associated with dopamine transmis-
sion. However, for many associations, it has been argued that extinction learn-
ing is not simply the “unlearning” of an association but new, discriminative
learning that associates different states of the world with new contingencies in
a more complex model. While there are clear demonstrations that extinction
can be new learning—most likely dependent on prefrontal cortical regions—it
perhaps should not be assumed that this is a universal mechanism. At least
at the level of the amygdala, molecules usually associated with the depoten-
tiation of synapses increase in activity during Pavlovian extinction, and these
molecules are also necessary for extinction to occur (Merlo et al. 2014). Thus,
it remains a possibility that different associations have fundamentally differ-
ent mechanisms of extinction. Regardless of mechanism, a defining feature of
intrusive thinking is that the underlying associations are relatively resistant to
extinction.
Reconsolidation
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Circuits That Mediate and Update Intrusive Thinking 19
inhibitor) can essentially erase the memory. Reconsolidation has been investi-
gated in the context of fear memories (Nader et al. 2000), where it was shown
to be dependent on protein synthesis in the basolateral amygdala. Likewise,
the basolateral amygdala is necessary for the reconsolidation of drug-related
associations that support place conditioning and conditioned reinforcement
(Milton et al. 2008a, b; Théberge et al. 2010). Interestingly, protein synthesis
in the nucleus accumbens core is also required for reconsolidation of asso-
ciations supporting conditioned drug place preference but is not required to
reconsolidate drug-cue associations that support conditioned reinforcement
(Théberge et al. 2010). The extent to which the disruption of reconsolidation in
one node of a motivational network can impinge on the function of other nodes
in that network is a question that has been surprisingly understudied. Despite
this, it has been hypothesized that reconsolidation could provide a mechanism
by which memories can be strengthened (Lee 2008), generalized (Vanvossen
et al. 2017), and integrated into wider memory networks (Hardt et al. 2010).
Therefore, if reconsolidation mechanisms were engaged upon reactivation of
an intrusive thought, it is possible that a form of mnemonic “positive feed-
back” could be established, by which a reactivated intrusive thought not only
becomes strengthened when it restabilizes, but generalizes and integrates with
other associative traces to lead to an increase in the number of cues and con-
texts that could trigger the intrusive thought. The consequent increase in the
likelihood of triggering the thought would potentially increase the likelihood
of subsequent reactivation, leading to further strengthening, generalization, in-
tegration, and so on.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
20 P. E. M. Phillips and A. L. Milton
Conclusions
The neural circuits that mediate and update intrusive thoughts are complex but
potentially tractable based on our current understanding of model-based and
model-free systems and their operation. It is important to appreciate, however,
that these circuits are not fixed and immutable, but rather it is likely that they
undergo repeated rounds of plasticity and metaplasticity, leading to imbalance
within the circuit. In this way, we hypothesize that an adaptive physiological
process, supported by functional neural circuitry, can become persistent, recur-
rent, and pathological.
Acknowledgments
This work was supported by NIH grants R01-DA039687, U01-AA024599, and P50-
MH106428-5877 to Paul Phillips, and U.K. Medical Research Council Programme
Grant MR/N02530X/1 to Amy Milton. Amy Milton is further supported by the Ferre-
ras-Willetts Fellowship in Neuroscience at Downing College, Cambridge.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
3
Corticostriatal Intrusions
Bernard W. Balleine
Abstract
The loop-like circuits which link the cortex and basal ganglia have been implicated in
a range of functions; most recently in the precursors to movement, including planning
and decision making. Damage to these circuits induced by various disease states have,
therefore, been heavily implicated in a range of symptoms, including intrusive involun-
tary thoughts and actions associated with, for example, neurodegenerative and psychi-
atric conditions as well as addictions of various kinds. This chapter focuses on recent
evidence of parallel circuits that mediate the distinct forms of control associated with
reflexive and volitional actions, and the interactions between these circuits in determin-
ing adaptive behavior. It discusses two kinds of interaction important for understanding
intrusive actions and thoughts: competitive interactions, whereby circuits controlling
volitional actions regulate reflexive or habitual responses, and cooperative processes
that allow the simulation of specific actions to become manifest in performance. It then
explores the role of information derived from predictive learning in action selection and
choice. The influence of such information is conveyed through a specific corticobasal
ganglia circuit, damage to which has been implicated in compulsive action. The evi-
dence considered generally suggests that intrusive thoughts and actions are the product
of an imbalance between corticobasal ganglia circuits rather than dysfunction in any
one circuit or its related control process.
Introduction
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
22 B. W. Balleine
networks that support these forms of action control. For the purposes of this
chapter, I will focus on two aspects of these networks that appear to have
important implications for future research into these issues. The first lies in
the relationship between the neural networks that mediate goal-directed and
habitual control: how they compete and cooperate to support our everyday
activities, and the consequences of failures in these interactive processes.
The second lies in the way an even more fundamental reflexive system me-
diating Pavlovian-conditioned reflexes interacts with goal-directed processes
to provoke the initiation and performance of otherwise volitional actions in
ways that can often appear maladaptive. Although both undoubtedly relate
to the habitual control of intrusions, neither wholly depends on such factors.
Both, however, highlight how actions and habits are integrated and, ulti-
mately appear to suggest that failures, when they occur, are failures in that
integrative process.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Corticostriatal Intrusions 23
Much of the evidence for distinct forms of action control has been derived
from the use of a battery of tests of goal-directed control through which the
sensitivity of animals, whether rodent or human, to changes in the action–
outcome relationship and in outcome value is evaluated (see Dickinson and
Balleine 1994; Balleine and Dickinson 1998; Balleine 2005). One such test,
the contingency degradation test, assesses sensitivity to the relationship be-
tween an action and its consequences by increasing the probability of access,
p, to a specific outcome, O, in the absence of an action, A, or p(O|noA). Given
a specific probability that an action earns an outcome (i.e., p(O|A)), p(O|noA)
can be increased until the outcome is equally probable if the action is per-
formed or not. At that point, the action is no longer causally related to the
outcome and, if the actor is sensitive to this, then the actor should no longer
perform the action (Hammond 1980). Increasing p(O|noA) beyond that point
means that the action prevents the outcome, which should reduce responding
even faster, something demonstrated through the use of omission schedules
(Davis and Bitterman 1971). Goal-directed action in humans and animals is
exquisitely sensitive to these manipulations of the action–outcome relation-
ship, largely because these manipulations alter the causal relationship between
these events (Dickinson 2012). As a consequence, not only is performance
affected, judgments as to how causal an action is, with respect to its specific
outcome, are similarly modified by these contingency manipulations (Shanks
and Dickinson 1991). Finally, these changes in contingency are highly selec-
tive; altering the relationship between one action and outcome does not affect
performance based on the causal relationship between other actions and their
outcomes (Balleine and Dickinson 1998).
The second commonly used test is called outcome devaluation, which as-
sesses sensitivity of performance to changes in the value of the outcome of an
action and is usually conducted without feedback (i.e., in extinction). Hence,
having learned a number of specific action–outcome relationships (A1 → O1,
A2 → O2, etc.), the value of one outcome can be altered, after which the sub-
ject is given a choice between the various actions. Goal-directed control re-
flects the ability to integrate the action–outcome relationship with the altered
value of the outcome to modify performance of the action. Again, considerable
evidence has demonstrated that goal-directed actions in both humans and ro-
dents are sensitive to these kinds of treatment (e.g., Balleine and Dickinson
1998; Balleine and O’Doherty 2010).
Neural Bases
Studies of goal-directed action have found evidence that the acquisition and
performance of such actions depend on the rich connections of the prefrontal
cortex (PFC), including the human ventromedial PFC and medial orbitofrontal
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
24 B. W. Balleine
cortices, with the striatum, specifically the caudate nucleus (Balleine and
O’Doherty 2010). The greater resolution allowed by rodent studies has signifi-
cantly refined this general picture. Thus, it is now clear that the homologous
region of the medial PFC in rats, the prelimbic cortex (area 32), is engaged
during the early acquisition of goal-directed actions (Hart and Balleine 2016).
Furthermore, whereas the input layers 2/3 contribute significantly to the con-
solidation of goal-directed learning, the output layers (particularly the intratel-
encephalic neurons in layer 5) are critical to this process. Activity in their
bilateral projection is necessary for learning-related plasticity in the posterior
dorsomedial striatum (one of their main targets) and thus for the acquisition of
goal-directed actions (Hart et al. 2018a, b).
Using the above tests, we demonstrated, some time ago, that rodents with
damage to the prelimbic cortex, the posterior dorsomedial, or the mediodorsal
thalamus show deficits in their sensitivity to contingency degradation and out-
come devaluation, and continued to respond as if neither the contingency nor
outcome value had changed. Evidence that these structures share a high degree
of interconnectivity led to the claim that they constitute the critical cortico-
striatal-thalamo-cortical loop through which goal-directed actions are encoded
(see Figure 3.1a; Balleine 2005). Importantly, this same insensitivity to shifts
in the action–outcome relationship and in outcome value is found in habitual
actions. Early in training, actions are sensitive to changes in contingency and
value, whereas with continued practice, performance naturally becomes in-
sensitive to these changes as action control shifts from the consequences of
an action to antecedent environmental stimuli with which the action becomes
associated (Dickinson et al. 1995, 1998). Chief among these stimuli is the con-
text in which the action is trained (Thrailkill and Bouton 2015).
At a neural level, the acquisition and performance of habits depends on a
corticostriatal circuit linking the sensorimotor regions of frontal cortex with
the putamen or dorsolateral striatum (Figure 3.1a). Again, treatments causing
degeneration to, or temporary inactivation of, structures in this circuit block
the acquisition of habit learning and render habitual actions goal directed; that
is, despite extensive training, they continue to show sensitivity both to shifts
in the action–outcome relationship and outcome value (Yin et al. 2004, 2006).
Generally, therefore, these findings have been interpreted as suggesting that
the goal-directed and habitual control of instrumental actions is a competi-
tive process: any reduction in goal-directed control increases the likelihood ac-
tions will be habitual, whereas any reduction in habitual control increases the
likelihood actions will be goal directed. Thus, inactivation of the dorsomedial
striatum immediately places actions under habitual control (Yin et al. 2005a),
whereas inactivation of the dorsolateral striatum appears immediately to ren-
der actions goal directed (Yin et al. 2006), as if these two processes are always
active and merely compete for control over performance (Balleine et al. 2009).
In fact, even under normal circumstances the goal-directed process can rap-
idly inhibit habitual control. This can be detected in our everyday activities. A
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
(a) Cognitive control Stimulus control
(c)
Goal-directed action Habitual action
S – R Xecute
MPC cortex SM
Senstive to changes Insenstive to changes
in their causal in their causal select
consequences consequences
Sensitive to changes Insensitive to changes
in goal value DMS striatum DLS in goal value
A – Ov
Rapidly acquired Slowly acquired
Flexibly deployed Inflexibly deployed evaluate
MD thalamus PO
Necessary for Impulsive and
inhibitory control perseverative
Habit memory Associative memory
SN midbrain GPi (d) …
S1 S2 Sn A S
(b)
R1 R2 … Rn
GDA Habit O
+ –
Arbitrator
R1 R2 Rn Evaluative memory
Motor output
Figure 3.1 Competition and cooperation between actions and habits. (a) Schematic of neural circuits mediating goal-directed and habitual ac-
tion control, as well as their primary characteristics: A loss of goal-directed control tends to yield dysregulated habitual actions, whereas a loss of
habitual control yields dysregulated goal-directed actions. (b) One view of the competitive processes in goal-directed action (GDA) and habitual
action control in which both compete for control of performance via some form of arbitration; habits emerge via arbitration but are heavily regu-
lated. (c) Illustration of the cooperation between the stimulus-mediated selection and action-mediated evaluation processes necessary to generate
action execution. Generally, stimuli (S) generate urges to respond (R) that initiate the retrieval of specific actions (A), their outcomes (O), and the
value (v) of those consequences. If negative, this evaluation would check that urge; if positive, the urge would translate into an executed action. (d)
This descriptive model can be further elaborated into an associative cybernetic model which views instrumental performance as the direct outcome
25
of cooperation between S-R and A-O associative processes, whose joint influence sums at the motor system to drive motor output.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
26 B. W. Balleine
useful example is to consider the effect of seeing a police car in the rearview
mirror while driving on the freeway: After a period of carefree, apparently
cognitively disconnected driving, the sight of the highway patrol causes a
very distinct change in our behavior. Do we carry on driving in such a care-
free manner? Not likely! Even if we are within the speed limit and are gener-
ally obeying the rules of the road, our vigilance would dramatically increase
and our driving would become far more deliberate. Thus, habitual control has
been suppressed. Likewise, rats behaving habitually will rapidly transition
to goal-directed control when the lever response is punished by the actual
delivery of an aversive or noxious outcome (Dickinson et al. 1983, 1995).
The rapidity of this adjustment is, however, severely curtailed by damage to
or inactivation of the dorsomedial striatum (Yin et al. 2005b; Furlong et al.
2017), a finding that is consistent with the argument that the return of control
to the goal-directed system is the source of the rapid suppression of habits in
a punishment situation (see Figure 3.1b).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Corticostriatal Intrusions 27
paired with drugs can render goal-directed action habitual in tests of outcome
devaluation (Ostlund et al. 2010; Corbit et al. 2014; Furlong et al. 2017).
Furthermore, actions in these contexts can persist when punished by the de-
livery of an aversive outcome (Furlong et al. 2018), something that we found
was linked to reduced activity in D1-expressing spiny projection neurons
(SPNs) in the posterior dorsomedial striatum (Furlong et al. 2017). There are
two populations of SPNs in the striatum that account for roughly 95% of the
neurons: direct path D1-expressing (dSPNs) and indirect path D2-expressing
SPNs (iSPNs). Whereas dSPNs tend to increase functional output, and hence
increase goal-directed control, iSPNs inhibit it (Gerfen and Surmeier 2011).
As such, a loss of dSPN activity should be expected to have the effects we
observed. In an attempt to rescue normal function, therefore, we attempted to
redress the relative balance between the activity of dSPNs and iSPNs by in-
hibiting iSPNs. How we achieved this was complex. The binding of dopamine
at D2 receptors on iSPNS inhibits the activity of these neurons and, as such, a
D2 antagonist might be expected to be sufficient. However, D2 receptors are
also expressed on multiple interneurons as well as iSPNs in the dorsal striatum,
reducing the selectivity of this manipulation. Importantly, however, adenosine
A2A receptors are only expressed on iSPNs in this region, and the inhibition
of these receptors increases the effects of D2 binding only on iSPNs (Tozzi et
al. 2007). We hypothesized, therefore, that local infusion of an A2A antagonist
would (a) increase D2 receptor activity, (b) reduce iSPN activity, (c) restore
the balance between dSPNs and iSPNs, and thus (d) allow the animals to exert
behavioral control over their instrumental performance. This is indeed what we
found: rats were now able to exert behavioral control over their habits and re-
duced performance in a punishment test to a similar degree whether they were
tested in a drug-paired or unpaired context (Furlong et al. 2017).
Similar deficits in contingency degradation and outcome devaluation have
been described in various psychiatric conditions linked to changes in the cir-
cuitry mediating goal-directed action control (Griffiths et al. 2014). We found,
for example, that the causal judgements of a cohort of youths diagnosed with
major depression were relatively insensitive to changes in the causal relation-
ship between action and outcome (Griffiths et al. 2015, 2016a). Furthermore,
reductions in causal awareness were correlated with size and shape changes of
the globus pallidus (GP) and midline thalamic structures in the basal ganglia
output circuit that feeds back to the cortex. Tractography confirmed the rela-
tionship between the dysfunctional area of GP and the striatum, on one hand,
and the midline thalamus, on the other (Griffiths et al. 2015). This suggests,
particularly given that subjects were at an early stage in illness progression,
that such changes may reflect the precursors of later prefrontal cortical and
corticostriatal deficits in depression, as has also been claimed by others with
regard to schizophrenia (e.g., Simpson et al. 2010). Indeed, in a recent study,
we found deficits in the flexibility of causal judgment in chronic schizophrenia
similar to that observed in major depression (Morris et al. 2018). Furthermore,
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
28 B. W. Balleine
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Corticostriatal Intrusions 29
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
30 B. W. Balleine
Although both prediction and control are necessary for adaptive behavior, dis-
cussions of the importance of predictive learning are usually confined to its
role in determining conditioned reflexes of one kind or another. There are,
however, multiple ways in which predictive learning can influence instru-
mental actions. Pavlovian predictors of important environmental events could
elicit conditioned reflexes congruent or incongruent with our actions; for ex-
ample, a stimulus paired with an aversive event could generate freezing, which
would be incongruent with active avoidance responses. Furthermore, there is
evidence that predictive learning can influence actions independently of the
conditioned reflexes predictors produce. Indeed, there is good evidence that a
stimulus that has become a reliable predictor of a valued outcome can enhance
the performance of actions associated with that outcome while leaving actions
associated with other rewarding outcomes unaffected. Furthermore, treatments
that abolish the predictive validity of such stimuli can abolish these effects on
instrumental performance without affecting their ability to evoke a conditioned
reflex (Delamater 1995). Over and above conditioned reflexes, therefore, pre-
dictive learning clearly provides information regarding forthcoming rewards
and punishers; if that information is important for adaptive behavior, its effects
will then be mediated by changes in instrumental action, via its effects on ac-
tions sufficiently flexible to be modified in response to that information and to
do so rapidly.
The influence of the information provided by predictive learning on in-
strumental action is typically studied in the laboratory using the Pavlovian-
instrumental transfer paradigm (for a recent review, see Cartoni et al. 2016). In
these experiments, subjects are exposed to a period of Pavlovian conditioning,
during which cues of various kinds are paired with specific, usually rewarding,
events (e.g., specific foods or fluids), after which they are trained to perform
distinct actions to earn those same food or fluid outcomes (Figure 3.2a). In a
typical rodent experiment, rats will be first given a period of predictive learn-
ing, during which they learn that stimuli S1 and S2 (e.g., tones or clickers)
predict the delivery of distinct outcomes, O1 and O2 (e.g., dry food pellets or
liquid sucrose): S1–O1 and S2–O2. Subsequently, they are trained to perform
two novel instrumental actions (A1 and A2) for these same outcomes. They
might be trained, for instance, to press one lever for the pellets and a second
lever for the sucrose: A1 → O1 and A2 → O2. In a final test, the previous
Pavlovian and instrumental phases are brought together to examine the influ-
ence of the former on the latter: rats must choose between A1 and A2 in the
presence of S1 and S2 (i.e., S1: R1 vs. R2, and S2: R1 vs. R2). Importantly, no
outcomes are delivered in this test phase, either after the stimuli or the actions.
Thus, the test provides an opportunity to observe how predictive learning influ-
ences instrumental performance directly. Typically, the stimuli cause the rats
to select and perform more vigorously the response previously associated with
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
(a) Predictive learning (c)
(Pavlovian conditioning)
mdT
Goal-directed learning
(instrumental conditioning)
Choice
(b) (d)
evaluates
selects Rn On
Medial
thalamic
Candate nuclei
Os v
VP
earns OFC
Rn On S
initiates
Figure 3.2 The influence of predictive learning on choice. (a) Design of a typical experiment used to study the influence of predictive learning
(Pavlovian conditioning) on instrumental choice performance, here in rodents. Rats are first exposed to two stimuli paired with distinct outcomes
and then trained to perform two actions for those outcomes before the influence of predictive learning on choice is assessed in an extinction test.
(b) Evidence (see text) suggests that the influence of predictive learning is mediated by the ability of a stimulus, S, to retrieve a specific outcome,
O (S – On), and so the specific action associated with that O (On – Rn). The outcome delivered as a consequence of Rn serves both as the goal of the
action and as a stimulus that selects the next action. (c) Schematic of the cortical-striatal-pallidal-thalamo-cortical circuits underlying goal-directed
action and the way predictive learning influences these actions. Note that the influence of stimuli on infralimbic and orbitofrontal cortex (IL/OFC)
activity is driven by the ability of nucleus accumbens shell (NAsh) and medial ventral pallidum (VPm) to inhibit mediodorsal thalamus (mdT),
and so disinhibit this input. The involvement of medial OFC and its connections with nucleus accumbens core (NAco) is crucial to the integration
of this circuit with the larger goal-directed circuit and its influence on performance through medial agranular cortex (M2). (d) Dysfunction in the
same cortical-striatal-pallidal-thalamo-cortical circuit in humans has been argued to underlie the compulsive actions associated with obsessive-
31
compulsive disorder; ventral tegmental area (VTA), caudate nucleus (CAUD) (after Modell et al. 1989).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
32 B. W. Balleine
the outcome predicted by the stimulus: given the scenarios described above,
S1: R1 > R2 and S2: R1 < R2. Very similar effects are observed in mouse and
human subjects to those observed in rats (Cartoni et al. 2016).
As intimated above, the influence of predictive stimuli on choice in the
transfer paradigm depends on their predictive status; degrading the contin-
gency between a stimulus and its specific outcome does not affect its ability
to elicit a conditioned response but completely abolishes its effects on choice
between actions. It is clear, therefore, that it is the information that the stimulus
provides with respect to a predicted outcome that is critical to the ability of
such stimuli to affect performance, rather than its ability to evoke the condi-
tioned response. Nor, as it turns out, is this effect due to the ability of the stimu-
lus to retrieve and activate goal-directed control generally via retrieval of the
action–outcome relationship and, subsequently, the value of the outcome. One
of the more interesting and telling effects in this literature is the finding that
devaluing the outcome predicted by the stimulus does not affect the ability of
that stimulus to influence choice (Rescorla 1994; Holland 2004). Although the
predictive validity of the stimulus is critical, the value of the outcome predicted
by that stimulus is not.
Finally, it is important to note that these effects of predictive learning on
choice between actions are mediated by both excitatory and inhibitory action–
outcome associations and their effects on performance. When a stimulus pre-
dicting a particular outcome elevates the performance of an action associated
with that outcome, it does so without affecting the performance of other actions
(Laurent and Balleine 2015). We have hypothesized that this is at least partly
due to the fact that, in goal-directed learning situations, actions can become as-
sociated with the outcome that they deliver (e.g., A1 → O1, A2 → O2) as well
as with the absence of outcomes that they do not deliver (i.e., A1 → no O2,
A2 → no O1). Hence, a stimulus that predicts a particular outcome is likely
to elevate the performance of actions associated with the predicted outcome,
but not the responses associated with the absence of that outcome; informa-
tion (e.g., S1 predicts O1) can elevate an action, R1, that delivers O1 but not
another action, R2, that predicts no O1 (Laurent and Balleine 2015). In sum-
mary, Pavlovian-instrumental transfer provides insight into the way predictive
learning affects instrumental performance. As a phenomenon, it reveals the
following:
1. Presenting cues that predict specific outcomes elevates the performance
of actions associated with those outcomes (without affecting those that
are not) by selecting those actions and increasing (or inhibiting) their
execution in an ongoing manner.
2. The ability of stimuli to produce these effects depends on how specifi-
cally they provide information about those outcomes.
3. The ability of these cues to provoke the performance of actions does
not depend on the value of the outcome with which they are associated.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Corticostriatal Intrusions 33
Of the theories advanced thus far to explain the effects of predictive learning
on instrumental performance, the one best supported suggests that the asso-
ciation between the stimulus and outcome, acquired during Pavlovian con-
ditioning, allows the stimulus to subsequently retrieve the outcome, thereby
retrieving (or inhibiting) the action associated with that outcome (Figure
3.2b; Balleine and Ostlund 2007). Any failure to inhibit competing actions
in the presence of a specific predictive stimulus would result in an unwanted,
intrusive action. This is strongly reminiscent of the occurrence of intrusive,
compulsive actions in various types of addiction as well as in multiple, se-
vere neuropsychiatric disorders, including Tourette syndrome (Leckman et al.
2010), grooming disorders (e.g., skin picking, trichotillomania; Chamberlain
et al. 2009), and obsessive-compulsive disorder (OCD) (Robbins et al. 2019).
Furthermore, not only does the superficial similarity of the behavioral influ-
ence of predictive learning on instrumental performance suggest this, there
is a very close similarity between the neural bases of Pavlovian-instrumental
transfer and the circuitry previously implicated in the compulsive actions as-
sociated with these conditions.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
34 B. W. Balleine
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Corticostriatal Intrusions 35
Over and above the relationship between the altered activity in various cortical-
dorsal-striatal circuits and the disease states described above, theories related
to the neural bases of addictive conditions and relapse as well as of OCD and
related conditions have implicated changes in the OFC-shell-VPm-mdT-OFC
loop mediating Pavlovian-instrumental transfer in those conditions (Fettes et
al. 2017). The role of this circuitry in OCD is particularly well documented,
especially the role of the vlOFC and mOFC and its reciprocal connections
with the mdT in this condition. Thus, for example, it has long been known that
the symptoms associated with OCD are ameliorated by surgical ablation of
either the OFC or of the midline and intralaminar thalamus, including the mdT.
Importantly, they are similarly ameliorated by thermal or gamma radio lesions
of the ventral anterior limb of the internal capsule (also called anterior capsu-
lotomy)—the region through which the white matter tracks pass containing
the bidirectional fibers which connect the thalamus and the OFC (Pepper et al.
2015). More recently, deep brain stimulation of these fiber pathways has been
reported to have a similar effect (reviewed in Greenberg et al. 2010).
Neuroimaging studies of OCD patients have implicated aspects of this cir-
cuitry in OCD, particularly the subregions of the OFC. Generally, the vlOFC
has been reported to be hypoactive (Remijnse et al. 2006) and the mOFC to be
hyperactive (Gillan et al. 2015) during tests that engage these regions; these
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
36 B. W. Balleine
findings are broadly consistent with other studies (e.g., Kahnt et al. 2012; Zald
et al. 2014). Two important findings from the animal literature suggest that these
differences in activity may be important for aspects of symptom formation. For
example, a study by Burguiere et al. (2013), which uses the SAPAP3 mouse
model of OCD, found that these mice show a persistent conditioned grooming
response when it was elicited by a cue that had been paired with a water drop to
the head (Burguiere et al. 2013). The authors hypothesized that this effect was
generated by a hypoactive vlOFC and thus sought to activate this region in an
attempt to reduce the excessive grooming response. They achieved this using
optogenetic stimulation of the vlOFC and found that the persistent grooming
was, indeed, inhibited by optogenetic stimulation of the lateral OFC neurons
projecting to the striatum. A similar study by Ahmari et al. (2013), optically
stimulated the terminals of mOFC cells expressing channelrhodopsin in the ven-
tromedial striatum, centering on the junction between the nucleus accumbens
core and shell. Stimulation of this pathway progressively increased grooming,
which persisted after the stimulation (Ahmari et al. 2013). As such it appears
that enhancing activation of accumbens projecting neurons in the vlOFC re-
duces compulsive grooming, whereas stimulating the mOFC projections to the
ventral striatum enhances compulsive grooming in line with the hypo- and hy-
peractive phenotypes of the vlOFC and mOFC in OCD, respectively.
Sometime ago, based on similar kinds of observations in OCD patients,
Modell et al. (1989) developed a circuitry model of OCD with striking simi-
larities to the circuitry involved in Pavlovian-instrumental transfer (Figure
3.2d). They argued that “the primary pathogenetic mechanism of OCD lies in
dysregulation of a basal ganglia/limbic striatal circuit that modulates neuronal
activity in and between posterior portions of the orbitofrontal cortex and the
associated medial thalamic nuclei” (Modell et al. 1989:32). Furthermore, they
proposed that specifically the compulsive symptoms of OCD are associated
with aberrant activity in a positive feedback loop in the reciprocally excitatory
frontothalamic neurons, due to the loss of inhibitory input from the ventro-
medial portions of the striatum. Further, they postulated that: “the net result
of ventral striatal activation is increased (inhibitory) pallidothalamic output:
essentially the loop transduces excitatory input from the vlOFC into inhibi-
tory output to the thalamus which serves to modulate activity of the frontotha-
lamic circuit by means of an interposed negative feedback loop” (Modell et al.
1989:32).
The same circuit is crucial to normal Pavlovian-instrumental transfer.
Damage to this circuit results in aberrant transfer; that is, it results in pre-
dictive learning eliciting both the performance of actions associated with the
outcomes predicted by those stimuli as well as actions that predict the absence
of those outcomes, which is the definition of an intrusive involuntary and mal-
adaptive movement. This hypothesis suggests, therefore, that the transfer ef-
fect and its attendant circuitry, which provides such an important translational
model of the cognitive control of actions, also provides a model of OCD in
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Corticostriatal Intrusions 37
which abnormality in this circuit results in the inhibition of the vlOFC and, by
extension, excitation of the mOFC.
Acknowledgments
Support for this work was provided by a Senior Principal Research Fellowship from
the National Health and Medical Research Council of Australia (GNT#1079561). The
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
38 B. W. Balleine
author thanks members of the Decision Neuroscience Lab at UNSW and attendees at
the ES Forum on Intrusive Thinking for the many discussions that have helped to shape
this work.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
4
Abstract
A critical challenge in the field of neuroscience as well as research into the neurobio-
logical basis of behavior has been to establish links between the cellular and biochemi-
cal processes within the brain and nervous system that occur during the mediation of
behavioral events. Developments over the last 10–15 years have provided several new
means to accelerate, advance, and dissect the specific mechanisms for brain function.
Developments across two realms of neuroscience and engineering have afforded re-
searchers, clinicians, and biologists advanced abilities to facilitate the dissection, ob-
servation, control, and perturbation of neural systems within intact, behaving animals.
These advances include electrical, optical, pharmacological, and specialized hardware
which allow for closed-loop interfaces to monitor and manipulate neural function. This
chapter explores how these recent developments have become integrated into our neu-
robiological tool chest. It describes current advanced approaches, and the limitations of
each, and explores future pathways toward even better technologies needed to dissect
the molecular, cellular, and circuit basis of behavior.
Introduction
The mammalian nervous system evolved over millions of years and contains a
heterogenous composition of networks and cells, which send messages to one
another to communicate information. This information is communicated by
a variety of signals: electrical, chemical, and anatomical (i.e., architectural).
These signals converge, amplify, or inhibit the flow of information to influ-
ence ultimately behavior in the organism. In many cases, specific central and
peripheral nervous system diseases are caused by dysfunctions in these brain
processes at molecular, cellular, and circuit-based levels. Some of these neu-
ropsychiatric disorders include, but are not limited to, addiction, pain, and
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
40 M. R. Bruchas
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
State-of-the-Art Toolbox for How Circuits Mediate Behavior 41
of cell types in the brain and spinal cord, along with the various scales at
which the nervous system operates, have made the design and implementation
of various neurotechnologies and interfaces challenging. Here I will describe
several technologies currently employed to observe and record neuronal ac-
tivity in awake or head-fixed animals: how the technology is currently be-
ing used, as well as future iterations of related methodologies. Devices that
interface with the nervous system must have the ability to decipher signals at
multiple scales as well as sufficiently interface with biological tissue in a flex-
ible, noninvasive manner.
Electrophysiological Methods
For nearly a century, the electrical properties of the brain and nervous sys-
tem have been empirically measured and recorded. Investigators have used
electrical probes to stimulate and mimic neuronal activity patterns in in-
vestigations of neural circuit functions during behavior. For many years,
researchers were severely limited by the “channel count” of the electro-
physiological probes for in vivo measures. This meant that we could not
sufficiently sample large populations of neurons in behaving animals. More
recently advanced materials engineering and manufacturing approaches
have, however, brought forth new technologies, including large Utah ar-
rays and Neuropixels (Jun et al. 2017). This latter technique integrates 960
recording locations within a 70 × 20 μm2 area and allows for single unit
action potentials to be isolated at very high resolution across multiple brain
regions. In addition, it reportedly allows for stable tracking of single neuron
activity over multiple days, thus allowing investigators to measure dynamic
changes within neural circuits that change over time. This method permits
significantly advanced throughput in neuronal recordings and provides the
ability, given their small size, for simultaneous recordings of activity to be
made over a wide range of brain regions.
Recent advances have also worked toward developing neural probes that
are softer and more flexible in their ability to interface with the brain. Most
neural probes are composed of hard surfaces, including metal, glass, and
silicon semiconductors; these materials can activate the brain’s immune re-
sponse and lead to severe lesion, inflammation, and cell death. Recently, an
injectable platform composed of mesh-based electronics, including 16 chan-
nels of platinum recording and stimulating electrodes, has become available
(Zhou et al. 2017). This device was designed in a flexible, open manner to
facilitate better integration with surrounding brain tissue and has the ability
to isolate local field potentials as well as single units. It can also be operated
chronically in the rodent brain for many months. This method provides ad-
ditional surface and material structure that can interact with the brain in a
noninvasive manner, while providing high-resolution actuation and observa-
tion data sets.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
42 M. R. Bruchas
Current Limitations
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
State-of-the-Art Toolbox for How Circuits Mediate Behavior 43
generated to provide powerful new abilities that further extend the function
and utility of the optical approach. These include channels with faster kinetics,
step-function properties, altered activation spectra (redshifted), and cellular lo-
calization (Yizhar et al. 2011; Klapoetke et al. 2014).
As these optogenetic stimulation tools became available, additional ap-
proaches were being developed to allow investigators to “silence neuronal
activity.” Here, the principal strategy was to develop photosensitive cation
channels which could act to hyperpolarize a neuron and thus significantly min-
imize and prevent that neuron from generating an action potential and “firing.”
These inhibitory opsins include pumps for protons hydrogen (H+) (called ar-
chaerhodopsin or bacteriopsin), sodium (Na+), and chloride (Cl–) (called halo-
rhodopsin) (Yizhar et al. 2011). Like ChR2, these proteins have been modified
to enhance function, expression, sensitivity, and wavelength, thus affording
investigators more advanced methods to manipulate neural circuits. The key
advantage to these inhibitory opsins is their ability to be harnessed for the
determination of how a genetically defined neuronal population is necessary
for a given behavioral event. The investigators can time lock activation of the
optical silencing method within a given group of neurons and observe the be-
havioral consequences of that particular manipulation in real time.
While the optical tools described above offer spatiotemporal and optical
control of neuronal activity through excitation or inhibition of given subsets
of neurons, due to the constraints of their binary impact and the fact that
their activation does not necessarily mimic naturalistic neuronal activity,
additional optogenetic approaches have been developed. These include, for
instance, methods for specifically manipulating cellular signaling, neuromod-
ulation, and gene expression. In particular, a newer approach was developed
whereby GPCR signaling (the primary mediator of neuromodulatory function
in the nervous system) can be mimicked in a cell type- and neural circuit-
specific manner. These chimeric optogenetic tools have been engineered us-
ing seven transmembrane-spanning opsins that contain the intracellular loops
and C-terminal tail of GPCRs, which are typically expressed within mamma-
lian neurons or glial cells. One of the first families of opto-XR receptors was
the adrenergic receptor system, whereby optically active beta-2 and alpha-1
adrenergic receptors were generated (Airan et al. 2009; Siuda et al. 2015).
Subsequently, a series of new opto-XRs, developed over the years, can acti-
vate a whole host of G-protein signaling pathways and neurons, including Gi,
Gq, and Gs among others (Spangler and Bruchas 2017).
Finally, several other photoactivatable proteins have recently been em-
ployed in cellular studies and are beginning to be used in vivo in brain tissue.
These optogenetic tools target the inhibition or activation of second mes-
senger cascades (for a review, see Wiegert et al. 2017). These tools regulate
downstream signaling using allosteric or proximity-based effects. They in-
corporate the use of flavoprotein domains, such as the light-oxygen-voltage
domain and cryptochromes. These flavoproteins generally fit into one of
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
44 M. R. Bruchas
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
State-of-the-Art Toolbox for How Circuits Mediate Behavior 45
Key Limitations
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
46 M. R. Bruchas
a single planer view. This approach is limited to a fixed neuronal number due
to laser power and coupling restraints; however, the approach suggests that
advanced optical holographic methods can be utilized to overcome the limits
of monolithic, synchronous optical manipulations.
The complexity of neural circuits in the mammalian nervous system has been
a daunting task to dissect. Canonical approaches have utilized in vivo elec-
trophysiological methods to record activity of neurons and ensembles within
discrete brain regions during behavioral tasks in awake, freely moving, or
head-fixed animals. This method, used for many decades, has provided neu-
roscientists with a rich framework to understand how various networks in the
brain respond under various behavioral states. While extracellular and multi-
dimensional (high channel count) recordings have been instrumental to our
understanding of neural circuits, they have been limited, for instance, by the
following factors:
• They lack genetic or cell-type identification.
• They are unable to track neurons across days and trials with confidence,
due to limitations of maintaining a single neuron during recording over
multiple sessions.
• Channel and cell count are limited by region and array size, and for
deep brain, significant issues arise due to lesioning of more dorsal
structures in an attempt to reach limbic structures.
Fortunately, several modern approaches have advanced our ability to detect
and visualize discrete neural circuits as well as to make claims about causality
of various cell types, circuits, and networks in mediating a particular behavior.
Although some of these approaches have only begun to be utilized widely in
the community, they are at the forefront of neural technology and are likely to
lead the field’s efforts in the coming years.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
State-of-the-Art Toolbox for How Circuits Mediate Behavior 47
signal into a fluorescent signal that can be measured using advanced micros-
copy approaches. This allows for a reliable proxy measure of action potential
firing patterns, along with synaptic calcium dynamics in real time.
The recent development of ultrasensitive protein calcium sensors,
GCaMP6.0 series, has been transformative for the field at large, because they
allow for very reliable detection of calcium transient activity and circuit activ-
ity in deeper structures when coupled with the advent and use of both cranial
window-based imaging and the gradient refractive index lens (GRIN), which
provides optical advantages for researchers using single-cell imaging methods
within deep brain limbic circuits. Cranial window-based imaging in multipho-
ton applications can provide larger fields of view, so that hundreds and even
thousands of neurons can be imaged in a single animal over multiple behav-
ioral sessions. New “mesoscope” microscopes are becoming available for this
very purpose and will greatly expand the field. The utilization of these new
biological and hardware-based imaging tools with fiber photometry as well as
single- and two-photon imaging permits reliable terminal field and single-cell
detection of calcium transient activity over single and multiple trials span-
ning days to weeks. Various hardware has been developed and optimized for
maximizing the types of behavioral experiments in which GCaMPs can be
used. Fiber photometry has gained substantial popularity in recent years due
to its relative ease of use. This method employs a simple fiber optic and pho-
tometer-based detection system and provides a computationally simple data
pipeline (Gunaydin et al. 2014) for measuring “bulk” calcium dynamics in a
given neuronal population, and ease of use in freely moving behavioral studies.
Further specificity of single-cell activity has been gained through the advent
of GRINs lens-containing mini-endoscopes (Ghosh et al. 2011; Barretto and
Schnitzer 2012); this allows for a less than 2 g microscope to be mounted to an
animal’s head, and a complementary metal oxide semiconductor image sensor
for high-speed detection of calcium transients in single cells. This mini-scope
approach allows for deep brain imaging during freely moving behavior, to-
gether with single-cell tracking over the course of multiple behavioral sessions
(Mukamel et al. 2009; Xia et al. 2017). Finally, two-photon imaging in head-
fixed, awake-behaving rodents allows for high-resolution long-term imaging
with either cranial windows or GRIN lens implants for deep brain applications.
This head-fixed two-photon imaging can be coupled with complex behavioral
tasks, including virtual reality systems and spherical treadmills with simulated
environments for increased complexity in both endoscopic and cranial win-
dow-based imaging platforms (Zhang et al. 2018; Jennings et al. 2019).
Additional tools are under development and being tested. These include an
expansion of the color range for calcium indicator protein sensors, such as red
fluorescent calcium indicators (RCaMP) and jRGECO (Akerboom et al. 2013).
The advantage of these additional sensors is that they will improve the imaging
depth within intact brain tissue, since near-infrared light scatters less through
tissue. The other advantage of these red indicators is the future ability to allow
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
48 M. R. Bruchas
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
State-of-the-Art Toolbox for How Circuits Mediate Behavior 49
neurotransmitters and modulators in real time during freely moving and head-
fixed behavior, there are still several limitations, and advances will be forth-
coming. Currently available sensors are mostly directed toward the detection
of small molecules, including fast transmitters, cholines, and monoamines.
While efforts are ongoing to use the same technology to detect neuropeptides,
tool development has been limited due to the high-affinity nature by which
neuropeptides bind to receptors and the inclusion of the cpGFP into the third
intracellular loop of the GPCR. This is the same region of the receptor that
dictates G-protein coupling and high affinity binding. There are some promis-
ing developing neuropeptide GPCR-based sensors on this front, none of which
have yet been published or validated in vivo; however, if neuropeptides can be
detected in a meaningful and genetically defined manner, it will open a host of
very exciting possibilities in how neuropeptides coordinate neuromodulatory
function within neural circuits that mediate a variety of behaviors (e.g., stress,
anxiety, fear, addiction, and reward seeking).
These new transmitter sensors could open new avenues and enable us to
address long-standing questions about neurons that co-release fast transmitters
(e.g., GABA and glutamate) while simultaneously releasing monoamines and
neuropeptides. We may be able to answer important questions about whether
neuropeptides encode specific information on their own or in conjunction with
specific fast transmitters under certain circumstances. Furthermore, these types
of tools coupled with imaging would allow us to expand our understanding
into the molecular and cellular basis of organization within neuromodulatory
and neurotransmitter circuits. Are peptides active at specific locations, released
in dendrodendritic form in some cases, or just released in mass in a volumet-
ric manner? Although these types of biosensors hold great promise, further
enhancement of their quantum yield (signal to noise) and sensors for other in-
tracellular signaling molecules (including cAMP, kinases, and other cascades)
will be needed for in vivo systems-level experiments in awake-behaving ani-
mal studies.
Conclusion
Optical tools provide unique methodologies in our quest to dissect neural cir-
cuits associated with behavior. Equipped with novel biological tools as well
as new, less restrictive hardware and/or systems with higher resolution for im-
aging activity within neural circuits, investigators have been able to resolve
specific spatiotemporal properties of discrete cell types, neurotransmitters, and
neuromodulatory pathways in real time during discrete behavioral events. The
challenges posed by these new methods include their invasiveness, their lack of
temporal resolution (particularly with GECIs), as well as their dynamic range.
Hardware limitations, in terms of the imaging window, pose limitations, for
instance, on data stream management. As richer, high-resolution data becomes
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
50 M. R. Bruchas
available through these new approaches, deciphering the data and utilizing new
computational approaches becomes more imperative. Computational models,
in turn, are needed to handle the new data, thus posing a future challenge.
The advent of these new technologies begs the question as to whether any
of the tools described above might eventually pave the way to a better under-
standing of intrusive thoughts in animal models, and whether these could be
applied in clinical translation. From the discussions at this Forum, it is clear
that there are a variety of possible uses for novel tools in establishing causality
and translation—provided that some of the limitations can be overcome. These
include using closed-loop sensing of neuronal activity (GCaMP or other) and
optogenetic (ChR2 or halorhodopsin equivalents) or pharmacological manipu-
lations in a wireless setting. Real-time sensing during established behaviors
defined to represent “intrusive thoughts” across species, alongside real-time
feedback with optogenetic and pharmacological control, would establish cau-
sality and mechanisms for intrusive thoughts, at least in one sense. For ex-
ample, deep brain stimulation has been widely used in clinical settings for a
variety of neurological and psychiatric diseases, yet it has not been used in a
closed-loop setting, whereby certain neuronal signatures, biomarkers, or other
measures would be detected followed by a closed-loop infusion of a drug or
optical/electrical stimulation.
The approaches outlined here, including optogenetic, chemogenetic, and
electrical perturbation, could be amenable to these ideas if we could (a) mea-
sure signatures of intrusive thoughts that span particular brain regions with
particular biomarkers and (b) overcome the limitations of expressing viruses
in the human brain. Recent developments in retinal research and clinical trials
with adeno-associated viruses, along with other viral delivery methods and
many new hardware developments, could assist translational approaches in
the future.
Collaboration between cross-disciplinary computational neuroscientists,
biologists, psychologists, behaviorists, and clinical psychiatrists is of para-
mount importance and needs continual encouragement. Notwithstanding these
challenges, the range of tools in the neuroscience toolbox continues to grow,
offering innovative ways to resolve specific pathways, networks, and behav-
iors with increased granularity. Future efforts require specific focus on cross-
species corroboration, computation, and analysis.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
5
Convergent Experimental
Systems for Dissecting
the Neurobiology of
Intrusive Thought
A Road Map
Abstract
Nonhuman experimental systems (also known as model organisms) are critical for un-
derstanding the neurobiology of intrusive thought. These model systems allow for the
ability to manipulate specific neurocircuits, neurotransmitters, neuromodulators, and
physiological and intracellular signaling events associated with behavioral markers that
may be linked to intrusive thought. They permit unparalleled control over the external
and genetic environments in ways and to degrees that are not possible in humans. Intru-
sive thought is an emergent property of multiple systems: emotional, cognitive, motor,
and autonomic/somatic. In an animal model, one can ask specific questions about these
systems and how they may be linked to, permit, or suppress intrusions. For example,
how are specific connections, neuromodulators, or cell types involved in each of these
systems, and how do they help form or maintain behaviors consistent with intrusive
thought? Are positive versus negative valences unbalanced? Are common systems
Group photos (top left to bottom right) Shannon Gourley, Antonello Bonci,
Suzanne Haber, Peter Kalivas, Amy Milton, Michael Bruchas, Shelly Flagel, Paul
Phillips, Shannon Gourley, Jeremy Seamans, Antonello Bonci, Suzanne Haber, Marina
Picciotto, Paul Phillips, Peter Kalivas, Amy Milton, Jeremy Seamans, Marina Picciotto,
Shannon Gourley and Antonello Bonci, Shelly Flagel, Michael Bruchas
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
54 S. L. Gourley et al.
hijacked by intrusive thought, agnostic to the valence or content of the thought? Resolv-
ing these issues could be transformative for the treatment of several neuropsychiatric
illnesses that are commonly characterized by intrusive thought. This chapter presents a
road map for studying the neural mechanisms underlying intrusive thought using non-
human experimental systems.
Introduction
Our first goal is to set forth principles by which we can capture aspects of
intrusions within the domain of experimental systems. Intrusive thought
has been defined as unwanted, unintended, conscious mental events lack-
ing control (Clark 2005). The aspects of this definition that we are best able
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems for Dissecting the Neurobiology of Intrusive Thought 55
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
56 S. L. Gourley et al.
1. Intrusive thought is, to some degree, conserved across rodent and pri-
mate species (and thus likely has some adaptive origins).
2. Corollaries and consequences of intrusive thought can be measured and
quantified in the absence of speech.
To the first point, one can imagine instances in which multiple types of
organisms would benefit from uninterrupted and intensive thought, such
as when the goal is to escape from a predator. However, an individual
must also be able to modify and shift focus when the situation changes
(e.g., when the threat has been resolved) and engage in other behaviors
that are more adaptive or otherwise suited to present and evolving con-
texts. A failure to inhibit intrusive thinking can distract from achieving
adaptive goals.
As shown in Figure 5.1, we conceptualize intrusive thought to be an
emergent property of multiple systems: emotional, cognitive, motor, and
autonomic/somatic (for discussion of the origin of intrusive thought in these
systems, see Roberts et al., this volume). We envision a world represen-
tation that contains these four coexisting elements, which homeostatically
analyze and validate environmental and intrinsic (thoughts) stimuli to elicit
adaptive behavior. Emotional and motivating content draw on circuitry in
the central zone of Figure 5.1. Intrusive events trigger a deviation from
the homeostatic condition; thoughts contain more excessive motivational
and attentional relevance to the individual than is appropriate for the envi-
ronment. In neuropsychiatric pathologies characterized in part by intrusive
thoughts, this deviation is associated with a loss of proper regulation of the
inner circuitry by the outer circuitry, as indicated in Figure 5.1 (for defini-
tions of typical vs. intrusive thoughts envisioned by the model in Figure 5.1,
see Table 5.2).
We envision that any given mental health disorder can coopt different do-
main hierarchies. Identifying these hierarchies could offer clues into the neu-
rocircuits that one might explore in investigating etiologies and developing
treatment strategies. For example, in disorders in which cognitive behavioral
therapy can be effective, such as OCD, the cognitive domain plays a significant
role in generating overall circuit feedback that restores homeostasis and con-
trol of the intrusions. We hypothesize that the distinct disorders or endophe-
notypes of disorders defined by DSM-5 have the order of domain dominance
shown in Table 5.3.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems for Dissecting the Neurobiology of Intrusive Thought 57
Cognitive
Cognitive Thalamo-
Thalamo- cortical
cortical
Autonomic
Somatic
Autonomic
Somatic
Keeping in mind the definitions, considerations, and concepts laid out above,
we can begin to develop a meaningful list of behavioral and other factors that
could be tractably measured in experimental systems (Table 5.4). For instance,
we could capitalize on the ability of an external stimulus to distract an experi-
mental subject from engaging in goal-directed actions:
• Will a rat in an operant-conditioning testing chamber respond for food
reinforcers even in the presence of an opioid-related cue?
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
58 S. L. Gourley et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems for Dissecting the Neurobiology of Intrusive Thought 59
• Alternatively, will the rat instead attend to the cue at the expense of
goal-directed food seeking?
• What differs, at a neurobiological level, between the rat who becomes
distracted and the rat who stays on task?
Similar to food-seeking behavior, drug-seeking behavior can have goal-di-
rected properties, but the important measure that one might wish to collect in
the context of intrusive thought is the degree to which it competes with a pre-
sumably more adaptive behavior, such as food seeking in a calorie-restricted
organism.
Some recently developed procedures (a) force organisms to arbitrate be-
tween food-seeking behaviors and the avoidance of aversive stimuli and then
(b) track the extinction of avoidance behavior in the absence of the stimulus
(Bravo-Rivera et al. 2015; Rodriguez-Romaguera et al. 2016). Others measure
the reaction of organisms to uncertainty (d’Angelo et al. 2014, 2017; Eagle et
al. 2014; Morein-Zamir et al. 2018). Both strategies could be used to investi-
gate mechanisms of intrusive thought, particularly when considered with fac-
tors such as individual differences or antecedents to intrusive thought, many of
which can be recapitulated in the laboratory (Table 5.5).
Another type of intrusion that could be recapitulated in nonhuman exper-
imental systems is the sense of incompletion of a task that looms until the
process is completed. One example in humans draws from obsessive hand-
washing in OCD: the notion that one’s hands must be washed is all consuming,
generating hyperarousal and stress until one washes their hands, resolving the
intrusion. Procedures in nonhuman experimental systems, such as persistent
Table 5.4 Examples of potential domain hierarchies involved in disorders containing
maladaptive intrusions. The interacting domains defined in Figure 5.1 may be associ-
ated with particular neuropsychiatric conditions to a greater or lesser degree.
Table 5.5 Antecedents to intrusive thought can be recapitulated in the laboratory. For
arousal, long-term events are historical events that give rise to vulnerabilities and resil-
iencies, whereas short-term events refer to triggering factors.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
60 S. L. Gourley et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems for Dissecting the Neurobiology of Intrusive Thought 61
Treatment Mechanisms
The recurrent nature of intrusive thoughts suggests that some type(s) of similarly
recurrent oscillatory or reverberating neural processes could be identified in ef-
fective models. These processes could then be exploited to expand understanding
into the etiology of recurrent thought, a strategy that we refer to as “decrypting
the ensemble.” This strategy is inspired, in part, by evidence that in subcortical
areas, the plasticity of conditioned fear-related behaviors is accompanied by tran-
sient, measurable changes in the expression of calcium-permeable and calcium-
impermeable α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA)
receptors (Clem and Huganir 2010; Rao-Ruiz et al. 2011). Additionally, changes
in the ratio of subtypes of N-methyl-D-aspartate (NMDA) receptors can deter-
mine whether a fear memory ensemble is stable or susceptible to strengthening
following plasticity (Holehonnur et al. 2016).
The specific content of intrusions can be common across a large number of
patients and interacts with the environment (e.g., an increase in the number of
patients reporting obsessions regarding contamination with acquired immune
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Table 5.6 Approaches to empirically investigate aspects of intrusive thought in nonhuman experimental systems. These approaches are not mu-
62
tually exclusive. Each approach is named and briefly described. Each strategy could be combined with stimuli meant to trigger or exacerbate an
intrusive thought and/or procedures that introduce conflict. The motivation to pursue a specific strategy is described (i.e., the strengths of a given
approach) and examples of each strategy provided. These lists are not exhaustive and some examples cross categories.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Table 5.6 (continued)
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
64 S. L. Gourley et al.
deficiency syndrome in the 1980s and 1990s). It is likely, therefore, that there
are particular concepts and associated neural ensembles that are recruited for
such intrusions. This phenomenon may reflect intrinsic differences in excit-
ability in specific neuronal populations, as is the case for amygdala neurons
which show high levels of cAMP response element binding protein (CREB)
phosphorylation (i.e., activation), which are subsequently more likely to be
recruited to fear-related neuronal ensembles following an aversive experience
than other neurons (Josselyn et al. 2001; Frankland and Josselyn 2014; Josselyn
and Frankland 2018). In other words, this subset of neurons with higher levels
of phosphorylated CREB are primed to respond to input. They therefore could
be an ensemble that could predispose an individual to an emergent (intrusive)
event. Similar phenomena have been described in the context of immediate
early gene expression (Suto et al. 2016; Whitaker et al. 2017), and these kinds
of strategies could be deployed formally to study intrusive thought.
Considerations
It is important to acknowledge multiple limitations inherent in nonhuman ex-
perimental systems as well as in the strategies that we propose. These include,
but are not limited to, the following:
1. There is a risk that phenomena unrelated to intrusive thoughts are
measured.
2. False negatives may arise, due to an inability to collect sufficient in-
formation (e.g., in cases of multiple interacting brain regions and/or
neuromodulators).
3. Manipulations of molecules/cells using many current tools may not be
physiologically relevant (see Bruchas, this volume).
4. In experimental organisms, the emotional content of any given manipu-
lation is difficult to measure conclusively.
Utilizing multiple strategies should help us overcome these challenges and
identify convergences.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems for Dissecting the Neurobiology of Intrusive Thought 65
A primary goal of this chapter is to offer a road map for future investigations
of intrusive thought. In this final section, we describe a selection of relevant
published and ongoing investigations and consider how these investigations
may be further developed to interrogate intrusive thought in nonhuman experi-
mental systems.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
66 S. L. Gourley et al.
interactions between these domains flow smoothly, with each taking a leader-
ship role in the appropriate situation but then returning to the status quo. For
example, an internal or external stimulus might invoke fear, overwhelming
the individual until either an escape (movement) is executed or the cognitive
control system takes over if the fear is unwarranted. Mediating between these
functional domains requires the complex integration of information across
them to resolve the situation. Intrusive thought can be considered to be a con-
dition in which this mediation fails. Identifying the circuitry that underlies the
integration of information processing across the different domains is a first
step in understanding how and where these areas communicate, necessary to
developing new therapeutic targets.
Karl Wernicke first recognized that connectivity of brain structures, rather
than their locations, was the central feature of higher-order cognitive functions
(Wernicke 1885/1994). Expanding on the idea, Geschwind suggested that this
involves a combination of functional localization and connectivity, leading to
the idea that the brain is comprised of complex, interrelated functional net-
works (Geschwind 1965a, b; Catani and Ffytche 2005). Functional imaging
studies and graph theory techniques moved the field forward, demonstrating
large-scale distributed networks and the existence of nodes and hubs (Sporns
2011). A node is an area that is connected locally or connected within a func-
tional system. A hub is a node of a network that has unusually high connectiv-
ity to other nodes, or degree centrality, and high connectivity to other hubs, or
eigenvalue centrality (van den Heuvel and Sporns 2013). Hubs are thought to
represent regions for integrating and distributing information from multiple
cortical regions. They likely play an important role in cross-functional compu-
tational tasks, such as integrating limbic, cognitive, and motor control calcula-
tions for decision making.
The rostral anterior cingulate cortex (rACC) is a good candidate for con-
taining hubs, because it sits at the connectional intersection of the emotion,
cognition, and executive control networks. Indeed, the entire rACC is consid-
ered a hub of the brain’s global network (Buckner et al. 2009). However, the
region is large, and inputs from different prefrontal cortical (PFC) functional
domains vary across it. These connections could represent information pro-
cessing sequentially across subregions (i.e., from valuation to cognition to ac-
tion). Alternatively, a hub could be embedded within the rACC that integrates
information across them. Consistent with the literature (Morecraft and Tanji
2009; Morecraft et al. 2012), mapping the distribution and relative strength
of frontal cortical inputs across rACC in nonhuman primates reveals that PFC
inputs to the rACC follow three general gradients:
1. Ventromedial PFC (vmPFC) and frontal pole inputs are strongest in
the ventral and rostral parts of the rACC, with decreasing strengths in
dorsal and caudal regions.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems for Dissecting the Neurobiology of Intrusive Thought 67
2. Frontal eye fields and premotor areas inputs are strongest in the dorsal
and caudal regions, decreasing in rostral and ventral rACC regions.
3. Ventrolateral PFC (vlPFC) and dorsolateral PFC (dlPFC) inputs peak
in a more central position.
One region embedded within these gradients, however, receives inputs from
unexpected additional areas. In addition to inputs from expected connections
from cognitive control areas, the dlPFC and vlPFC, this region is also con-
nected with regions that are part of the emotional system, the orbitofrontal
cortex (OFC) and vmPFC, as well as with the sensorimotor system, the frontal
eye fields (Tang et al. 2019). Thus, this connectional hub within the rACC is
in a position to integrate information across emotional, cognitive, and senso-
rimotor systems. It is perhaps unsurprising, then, that both PTSD and major
depressive disorder show treatment response in an area in close proximity to
the rACC hub (Mayberg et al. 1997; Pizzagalli 2011; Chakrabarty et al. 2016).
The striatum is also an important structure for integrating and distributing
information. Although the striatum is classically divided into limbic, cognitive,
and motor regions, embedded within this general topography, terminals from
different frontal cortical areas interface in the rostral striatum, positioning it
optimally to contain hubs (Haber et al. 2006; Averbeck et al. 2014). Indeed, in
a specific location within the rostral caudate nucleus, terminal zones from the
inferior parietal lobule, an area important for perception, converge not only
with those from the dlPFC and vlPFC, as expected (Cavada and Goldman-
Rakic 1991; Yeterian and Pandya 1993), but also with projections from the
OFC and rACC. Thus, similar to the rACC, this hub combines inputs from
several functional domains. The connections of the rACC and striatal hubs are
examples of highly integrative, cross-functional regions with distinct combi-
national inputs that provide the anatomical substrate in which computations
about motivation, internal states, cognition, perception, and motor control are
linked to mediate adaptive behaviors based on the interaction of these func-
tions. Disconnection of these hubs will likely result in an imbalance between
goal-directed control, emotion, and higher cognition, and thus play a key role
in maintaining intrusive thoughts.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
68 S. L. Gourley et al.
systems to generate operant responses for rewards such as food or juice. Then,
the experimenter modifies the likelihood that a given behavior will be rein-
forced, and organisms must update learned action–outcome associations to
modify their responding optimally. In a series of investigations, inactivation of
the vlOFC blocked the ability of mice to update response strategies (Gourley
et al. 2013a; Zimmermann et al. 2017, 2018; Whyte et al. 2019), consistent
with evidence that certain vlOFC neurons represent outcome-related memories
(Namboodiri et al. 2019).
The vlOFC interacts with aspects of the dorsal striatum, a key constitu-
ent of goal-directed action, to coordinate action–outcome response flexibil-
ity (Gourley et al. 2013a; Gremel and Costa 2013). Meanwhile, the use of
viral-mediated gene silencing and behavioral pharmacological strategies has
revealed several essential molecular factors within the vlOFC that optimize
its function. These factors include, but are likely not limited to, brain-derived
neurotrophic factor (Gourley et al. 2013a; Zimmermann et al. 2017; Pitts et
al. 2020) and its high-affinity receptor trkB (Pitts et al. 2018, 2020), Abl2 ki-
nase (DePoy et al. 2017), GABAAα1 receptor subunits (Swanson et al. 2015),
fragile X mental retardation protein (Whyte et al. 2019), and developmental
expression of integrin receptors (DePoy et al. 2019). These investigations
provide overwhelming evidence that the vlOFC is necessary for behavioral
switching, and they potentially shed light on molecular factors that are dis-
rupted when intrusive thoughts interfere with behavioral flexibility essential to
day-to-day function.
One common factor linking all of these proteins is that they regulate the
stability or turnover of dendritic spines, the primary sites of excitatory plas-
ticity in the brain. Whyte et al. (2019) revealed that updating expectations
regarding whether an action was likely to be rewarded reduced thin-type
dendritic spines, considered immature, on excitatory vlOFC neurons in mice.
Meanwhile, the proportion of mushroom-shaped spines, considered mature
and likely containing synapses, increased, potentially solidifying newly mod-
ified action–outcome associations to optimize future decision making and be-
havioral flexibility.
One function ascribed to the OFC as a whole is the updating of expecta-
tions, particularly under ambiguous circumstances; by extension, unbalancing
these connections via spine loss or inappropriate excitatory plasticity could
render expectations ambiguous and thereby vulnerable to intrusion by compet-
ing impulses. Consistent with these notions, exposure to cocaine (Gourley et
al. 2012a; DePoy et al. 2017; Pitts et al. 2020) and stress hormones eliminates
dendritic spines in the vlOFC, and identical procedures cause failures in the
action–outcome updating of stress hormones (Gourley et al. 2012b, 2013b;
Barfield et al. 2017; Barfield and Gourley 2019). Further, drugs that improve
behavioral updating appear to recruit local cytoskeletal regulatory systems
(DePoy et al. 2017). As a final note, artificially stimulating excitatory neurons
in this region also causes failures in action–outcome updating (Hinton et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems for Dissecting the Neurobiology of Intrusive Thought 69
2019), potentially by activating circuits associated with OCD. This idea is dis-
cussed at length by Balleine (this volume). Understanding the conditions under
which specific vlOFC connections are stimulated or quiescent could shed light
on how thoughts and actions can fail to be updated and become intrusive, and
how one recovers from their intrusion by switching cognitive strategies or be-
haviors, maintaining adaptive flexibility between emotional, cognitive, motor,
or somatic domains (Figure 5.1).
Unfortunately, cellular heterogeneity within several brain regions pertinent
to this discussion remains undefined. For instance, even within the striatum,
where cells subtypes can readily be distinguished based on dopamine recep-
tor constituents, dopamine D1 and D2 receptor-mediated neuronal ensemble
responses to cues and rewards over temporal dimensions, and as a function
of experience, remain opaque (for review, see Castro and Bruchas 2019).
Determining the contribution of each cell to learning, memory, stressor, and
drug reactivity, for instance, has previously been studied using in vivo physi-
ological approaches, yet the process of defining circuit- and cell-type specifics
in time and space is still in its infancy. As such, comprehensively understand-
ing the neurobiological bases of neuronal ensembles is a critical step if we are
to dissect and understand the aberrant patterns (signatures) by which intrusive
events occur.
Arousal Systems
Intrusive thinking includes arousal, and deviations from typical arousal states
can predispose one to, or acutely trigger, intrusive thoughts, a notion empha-
sized in Table 5.5. As such, understanding the mechanisms of arousal pres-
ents a point of entry into understanding intrusive thought itself. Within the
framework that arousal can decrease the threshold for permeation of intrusive
thoughts, we might consider the ability of acetylcholine (ACh) release elicited
by salient stimuli to alter the strength of signaling in thalamo-cortico-thalamic
loops, both acutely and persistently, via synaptic potentiation (Aramakis et
al. 2000; Kawai et al. 2007). This ACh-mediated elevation in activity may
alter the threshold for transmission of sensory information from subcortical
to cortical structures. The directionality of this signaling can vary across de-
velopment, with different ACh receptors mediating increases or decreases in
the transmission of sensory information (Aramakis et al. 2000; Heath and
Picciotto 2009).
Although we recognize that hallucinations may not reach a formal defini-
tion of intrusive thoughts, it could be useful to evaluate the particular circuits
for which we have direct evidence of a causal relationship with this perceived,
maladaptive mental event. Pharmacological studies (Warburton et al. 1985;
Fisher 1991), as well as evaluations of patients with loss of cholinergic neu-
rons (Dauwan et al. 2018), reveal that blocking muscarinic ACh receptors or
decreasing ACh levels in patients with Lewy body dementia (Tsunoda et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
70 S. L. Gourley et al.
2018; Dudley et al. 2019) results in hallucinations. One proposal is that loss of
either nicotinic or muscarinic activity in corticothalamic circuits may underlie
these hallucinations (Esmaeeli et al. 2019). Another suggestion is that cortical
ACh increases the signal-to-noise ratio of perceived events, and that “mus-
carinic receptor activation in the cortex is involved in confining the contents
of the discrete self-reported conscious ‘stream’ ” (Perry and Perry 1995:240).
When cholinergic input to the cortex is lost, irrelevant sensory information
normally confined to subcortical circuits enters conscious awareness (Perry
and Perry 1995), and hallucinations result from “a failure of the metacognitive
skills involved in discriminating between self-generated and external sources
of information” (Kumar et al. 2009:119).
An additional set of studies suggests that there is a pervasive increase in
ACh levels throughout the brain in patients who are actively depressed: for
unipolar depression, see Saricicek et al. (2012); for bipolar depression, see
Hannestad et al. (2013). Elevated ACh may be a risk factor for depression
since remitted patients have intermediate ACh levels between actively de-
pressed individuals and healthy controls (Saricicek et al. 2012), as measured
by competition with a cholinergic ligand and validated by within-subject
challenge with a cholinesterase blocker (Esterlis et al. 2013). Relatedly, ACh
is a critical mediator of arousal and rapid eye movement sleep (Ma et al.
2018). At baseline, ACh input to the basolateral amygdala is very high, and
tonic activity of the cholinergic system can thus control both the level of
arousal of stress-related systems and the likelihood that a stressful event will
activate the basolateral amygdala (Picciotto et al. 2012). Currently, there is
no information on whether this increase in ACh levels is associated with
intrusive thoughts (e.g., rumination in depression), but this topic could guide
future experiments.
Manipulations and measurements of both cholinergic signaling and circuits
modulated by its receptors may represent a cross-species neurobiological ap-
proach ripe for translational evaluation. One consideration is that intrusive
thoughts can be represented in experimental settings by regulation of arousal
states along a multidimensional continuum. A possible dimension along this
continuum includes asynchronous, decoupled activity of the filter or gain do-
main which prohibits “normal” function in a given circuit, steering an organ-
ism toward a hyperaroused state. One particular example of this phenomenon
is found in the locus coeruleus, which projects broadly throughout the brain,
and its activity (tonic vs. phasic) is dictated by salience, context, and stress re-
sponsivity. The ability of the locus coeruleus noradrenergic system to dissoci-
ate attention signals from stressful ones depends on which inhibitory filters are
engaged. Along the temporal dimension, intrusive thought may have the effect
of dysregulating the inhibitory gain signal (typically regulated by neuromodu-
lators such as neuropeptides, monoamines, and steroids), thereby producing an
unwanted hyperaroused state.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems for Dissecting the Neurobiology of Intrusive Thought 71
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
72 S. L. Gourley et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems for Dissecting the Neurobiology of Intrusive Thought 73
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
74 S. L. Gourley et al.
(CS) encodes the incentive value of the cue, not the predictive value (Flagel et
al. 2011b). Relative to goal trackers, sign trackers show greater engagement of
the cortico-thalamic-striatal “motive circuit” in response to a food cue (Flagel
et al. 2011a). Within this circuit, the paraventricular nucleus of the thalamus
(PVT) has emerged as a critical regulator of individual differences in cue-moti-
vated behaviors (Haight and Flagel 2014; Haight et al. 2015; Kuhn et al. 2018).
The PVT is a midline thalamic nucleus ideally located to integrate cog-
nitive, emotional, and arousal information from various areas of the brain
and, in turn, to guide motivated behaviors (Kelley et al. 2005; Kirouac 2015).
Specifically, the PVT receives dense input from the PFC, as well as subcortical
areas, including brainstem nuclei such as the dorsal raphe and locus coeruleus
and other areas such as the lateral hypothalamus and amygdala. The PVT then
integrates this information and sends reciprocal output to some of the same
regions, but also has dense glutamatergic projections to the shell of the NAc.
In fact, the PVT can regulate dopamine release in the NAc, even in the absence
of the ventral tegmental area (Parsons et al. 2007).
Within the context of the sign-tracker/goal-tracker animal model, neu-
rons projecting from the prelimbic PFC to the PVT appear to encode the
predictive value of reward cues, whereas subcortical systems surrounding
the PVT encode the incentive value. Specifically, sign-tracking behavior is
thought to result from hyperactivity of neurons projecting from the lateral
hypothalamus to the PVT, and those projecting from the PVT to the NAc
(Haight et al. 2017). The working hypothesis, therefore, is that cognitive rep-
resentation, or the predictive value of the reward cue, is encoded in prelimbic
PVC-PVT projecting neurons, and that this top-down process predominates
in goal trackers. In sign trackers, however, where incentive learning prevails,
subcortical processes are able to override this top-down mechanism. Thus,
the PVT appears to act as a fulcrum between top-down cortical processes and
bottom-up subcortical processes, and an imbalance between these processes
may result in aberrant or psychopathological behavior. It is also intriguing, in
light of our discussion of ACh systems above, that sign-tracking behavior in
rats is associated with poor attentional control, mediated by an unresponsive
basal forebrain cholinergic system (Kucinski et al. 2018). The neurobehav-
ioral endophenotype of sign trackers may capture antecedents that predis-
pose an individual to intrusive thoughts. For example, sign trackers appear
to have an inherent imbalance between emotional and cognitive domains
(with the PVT acting as a fulcrum between the two domains; see Figure 5.1),
and this imbalance renders them more susceptible to behavioral control by
intrusive experiences.
The concepts outlined above provide a number of avenues for future research.
For example, the idea that we can capture and decode the neuronal ensembles
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems for Dissecting the Neurobiology of Intrusive Thought 75
Conclusions
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
76 S. L. Gourley et al.
are represented in hubs throughout the brain. When the neural choreography
between these hubs and their corresponding nodes becomes disrupted, there
is a loss of homeostatic and/or cognitive control which leads to maladaptive
thoughts and inappropriate behaviors. In the laboratory, with careful experi-
mental design and multidimensional levels of analyses, we can model this loss
of control on both behavioral and neural levels. Here, we provided a road map
that illustrates multiple routes by which different approaches can be used in
combination to expand our understanding of intrusive thought. This road map
is not proscriptive; rather, we hope that it will serve as a foundation for a novel
avenue of preclinical research to advance our knowledge and ultimately lead
to more effective therapies for a number of psychiatric illnesses that are char-
acterized by intrusive thoughts.
Acknowledgments
The authors wish to acknowledge support from the United States National Institutes of
Health (grant numbers DA014242, DA044297, MH117103, MH007681, MH100023,
OD011132, DA039687, AA024599, MH106428-5877, DA033396, MH111520,
MH112355, MH045573, and MH106435).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
6
Abstract
This chapter reviews different methods that can be used to examine and understand
intrusive thought, beginning with behavioral methods. Common among these are
self-report and diary measures of the experience, duration, and intensity of intrusive
thoughts as well as self-reports of the difficulty in controlling such thoughts. These
questionnaires, for the most part, have been tailored to the types of intrusions specific to
a given psychiatric syndrome (e.g., flashbacks in posttraumatic stress disorder, thoughts
of contamination in obsessive-compulsive disorder), which highlights the need to cre-
ate a transdiagnostic self-report measure. Another common behavioral paradigm is to
investigate intrusions after individuals are exposed to traumatic material, through a
symptom provocation paradigm in individuals who have experienced trauma or an ana-
log trauma (e.g., viewing a disturbing movie). Other behavioral paradigms, such as the
Think/No-Think paradigm, specifically examine mechanisms of memory retrieval and
suppression often thought to be disrupted in posttraumatic stress disorder.
Thereafter, it addresses paradigms for examining the neural mechanisms associ-
ated with intrusive thoughts. These approaches primarily couple behavioral techniques
or paradigms with functional magnetic resonance imaging or electroencephalographic
(EEG/ERP) methods. In addition to providing insights into the neural mechanisms that
may underlie intrusive thoughts, these approaches may provide additional information
regarding cognitive mechanisms, such as discerning whether memories are being sup-
pressed or replaced. Discussion concludes by examining emerging approaches to the
study of intrusive thinking. A main challenge is to find a method to verify that intrusive
thoughts have indeed occurred. New paradigms that combine neuroimaging techniques
with computational methods drawn from machine learning offer promise, as do tech-
niques which allow intrusive thought processes to be examined as they occur during
more naturalistic processing (e.g., watching a film).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
80 M. T. Banich
Introduction
The question of how one can measure intrusive thoughts (i.e., thoughts that
“pop up” into consciousness in a seemingly uncontrolled manner) is difficult
to address, especially in a laboratory setting. It can be challenging to capture
and to verify their occurrence by measures other than self-report. Moreover,
traditional experimental measures used to explore mental processes (e.g., those
that determine reaction time and/or errors) are not applicable. Despite these
challenges, a number of approaches, both behavioral and biological, have
been employed. In this chapter, I review those methods, discuss some new
approaches, and suggest ways that the knowledge from other related arenas
of inquiry might be used to inform potential novel approaches to this difficult
question.
The need for scientists to have methodological approaches that will enable
an understanding of the basic cognitive and neural processes underlying in-
trusive thinking is apparent when one considers that intrusive thoughts are
ubiquitous across a large number of psychiatric disorders. Although intrusive
thoughts are generally associated with posttraumatic stress disorder (PTSD)
and are captured by the intrusions cluster of diagnostic criteria (American
Psychiatric Association 2013), they occur in many psychiatric disorders: de-
pression is characterized by intrusive negative thoughts and memory (Newby
and Moulds 2011), anxious apprehension (i.e., worry) by repetitive intrusive
concerns about future negative events (e.g., Fox et al. 2015), obsessive-com-
pulsive disorder (OCD) by thoughts of contamination and/or harm to self or
others (Bouvard et al. 2017), and schizophrenia by intrusions of semantic and
sensory information (Elua et al. 2012). This commonality raises the possibility
that intrusive thought across disorders may have a common underlying neural
circuitry (Kalivas and Kalivas 2016). What tends to vary somewhat across dis-
orders is the content of those intrusive thoughts (e.g., negative attributions of
the self in depression vs. concerns about potential dangerous future outcomes
in anxiety). In addition, intrusive thoughts in disorders such as PTSD are gen-
erally thought to be more sensorially based and of shorter duration, compared
to ruminative thoughts associated with depression and worry, which tend to
be more cognitive in nature, of longer duration, and recurrent (Speckens et
al. 2007). Nonetheless, what all these types of intrusive thoughts share is that
they appear to impinge upon consciousness in a somewhat uncontrolled man-
ner. Hence, my focus here is on methods for examining the complete range
of types of intrusive thoughts. To date, however, the majority of the work us-
ing methods to examine intrusive thought has focused mainly on individuals
with PTSD, OCD, and/or nonclinical populations. Thus, there is a much larger
range of individuals with psychiatric disorders to whom such techniques might
be applied.
Here, I will consider a number of different ways in which intrusive thoughts
have been examined. I begin with a discussion of behavioral paradigms, from
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Intrusive Thinking in Vivo and in the Context of Brain Mechanisms 81
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
82 M. T. Banich
Table 6.1 Examples of typical self-report measures of intrusive thoughts and the abil-
ity to control thought. Typical self-report measures of intrusive thoughts and the ability
to control thought. Example items from these measures are shown in italics.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Intrusive Thinking in Vivo and in the Context of Brain Mechanisms 83
These types of self-report measures have advantages: they are typically short,
easily administered, generally well normed, can be used with both clinical and
nonclinical populations, and scores derived from them can be used as covari-
ates in adjunct analyses. In terms of limitations, they require metacognitive
abilities related to self-awareness and self-evaluation on the part of the respon-
dent, which may be compromised in individuals with more severe psychiatric
disorders. Perhaps most glaringly, however, is the fact that they can be narrow
in scope, as most were designed to address a specific psychiatric disorder. As
such, they tend to examine the types of processes (e.g., punctate vs. continu-
ous) and specific topics of intrusive thoughts that characterize a given psychi-
atric disorder.
Hence, a questionnaire on intrusive thoughts and their control is needed
that could be used more generally across individuals with a variety of clinical
disorders, as well as with individuals who do not meet clinical criteria. There
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
84 M. T. Banich
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Intrusive Thinking in Vivo and in the Context of Brain Mechanisms 85
intrusions can then be examined within the context of the laboratory or via
a diary of intrusions for some specified time (e.g., one week) after exposure.
As discussed by Visser et al. (2018), a variety of points in memory pro-
cessing could be disrupted by or associated with intrusive thoughts: from the
original attention to and encoding of information to the access and retrieval of
memories. Procedures designed to engender intrusive thoughts can be com-
bined (a) to analyze the effects of other variables or manipulations at distinct
time points (before, during, or after exposure) and (b) to examine how differ-
ent processes (e.g., attention/focus, encoding, and recall) influence intrusions.
Manipulations implemented before exposure include having individuals recall
memories that induce high levels of self-efficacy (Krans et al. 2018) to put the
focus on themselves, or playing a distracting video game to place the focus
on something else (James et al. 2016b). Manipulations implemented during
viewing of the material include varying the cognitive load during the task or
instructions that lead to hyperarousal via hyperventilation (Nixon et al. 2007).
Manipulations after exposure have included varying instructions on how to
deal with the intrusions, such as rumination (“How can I drive again without
thinking about what could happen?”), integration so as to distinguish the video
from the person’s own nontraumatic experiences (“Think about your own driv-
ing experiences”), or distraction (“Try to recall as many African countries as
you can think of”) (e.g., Zetsche et al. 2009; Horsch et al. 2017).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
86 M. T. Banich
The vast majority of studies that examine intrusive thoughts do so from the
perspective of long-term memory formation and retrieval, which is particularly
appropriate when intrusive thought is driven by a specific event or circumstance,
such as occurs in PTSD. Far less work has focused, however, on mechanisms
related to intrusive thoughts, especially those not linked to a specific event, in
terms of how they get “stuck” in working memory and current consciousness.
Symptom provocation and analog trauma are not well suited to examining the
nature of recurrent, ruminative, and cognitively based intrusive patterns of think-
ing, which are typical in depression or worry (anxious apprehension) but cannot
be specifically linked to a particular point in time nor to a particular set of pro-
voking stimuli. For example, depressive intrusive thoughts often focus on how
one could “solve” the issues that lead to distress and negative affect. From think-
ing about social interactions with others to self-reflection on actions taken to an
analysis of one’s internal mood states, the topical range tends to be larger than,
for example, thoughts of contamination in OCD.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Intrusive Thinking in Vivo and in the Context of Brain Mechanisms 87
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
88 M. T. Banich
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Intrusive Thinking in Vivo and in the Context of Brain Mechanisms 89
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
90 M. T. Banich
The advantage of capturing brain processes associated with both the engender-
ing and controlling of intrusive thoughts is that they can provide more informa-
tion than a simple behavioral reaction time (i.e., button press) or retrospective
report. One must be cautious in inferring the engagement of cognitive pro-
cesses from patterns of brain activation, even within the context of the broad
set of knowledge regarding the neural circuitry underlying memory processes.
Nonetheless, these patterns provide insight into what aspects of memory pro-
cessing are disrupted during intrusive disorders, or whether control mecha-
nisms are intact but mainly engaged at inappropriate times. A disadvantage of
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Intrusive Thinking in Vivo and in the Context of Brain Mechanisms 91
such approaches is that these neural metrics often require multiple trials for
signal averaging; thus, the frequency of intrusions poses a potential limitation.
Although they may readily occur, with increasing practice at suppression, they
tend to become less frequent (Hellerstedt et al. 2016), which may limit the
amount of data that can be collected.
The studies discussed above that used fMRI focused primarily on brain re-
gions that become active during an intrusion or during the attempt to control
an intrusion. Additional metrics, however, should be examined to see whether
they can provide a distinct window into these processes. For instance, activa-
tion within cognitive control (e.g., dorsolateral prefrontal cortex) and mem-
ory-related regions (e.g., the hippocampus) has been implicated in suppressing
memory retrieval, as has the connectivity between these regions (e.g., Depue
et al. 2007; Benoit et al. 2015). Connectivity patterns could be examined us-
ing independent component analysis, which reveals groupings of brain regions
whose activity follows a similar temporal time course during the suppression
of a thought, as compared to other processes, such as visual imagery (Aso et
al. 2016).
A variety of electromagnetic techniques can be applied to studying intrusive
thoughts. These may focus on specific ERP components, such as the parietal
old/new component, which occurs approximately 50–80 ms after stimulus pre-
sentation and is thought to be an index of memory retrieval (Rugg and Curran
2007). Such measures could be combined with measures of neural oscillations,
recorded from the scalp (e.g., Depue et al. 2013) or via intracranial record-
ings in patients undergoing surgery for epilepsy (Oehrn et al. 2018), to pro-
vide information on the control of such retrieval. Magnetoencephaology has
been used to examine downregulation of sensory aspects of long-term memory
in the gamma band (70–120 Hz) in traumatized refugees (Waldhauser et al.
2018). Optical imaging methods (e.g., functional near infrared spectroscopy,
which provides information on both the location and time course of activation)
have been used to examine brain activation in individuals with PTSD during
symptom provocation (Gramlich et al. 2017) as well as in individuals high in
rumination during stress (Rosenbaum et al. 2018).
All of these methods record or otherwise observe the nature of brain ac-
tivation associated with memory retrieval or control processes. In contrast,
current work that focuses on using brain stimulation techniques (e.g., transcra-
nial magnetic stimulation or transcranial direct current stimulation) to alter
intrusive thoughts (i.e., to induce or disrupt them) is still preliminary. In one
study, brain stimulation of prefrontal regions and the underlying white matter
in three patients about to undergo surgery for epilepsy was found to induce
intrusive thoughts. For example, when stimulated one patient reported: “The
stimulation induces the disappearance of the word in my mind and replaces
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
92 M. T. Banich
it with something else” (Popa et al. 2016:3). Another reported that he had “a
thought that seems to come from nowhere” (Popa et al. 2016:4). To the best
of my knowledge, brain stimulation techniques have not been used to disrupt
thought. In addition, such methods may provide other insights into intrusive
thoughts. For example, aspects of intrusive memories in PTSD tend to be over-
generalized (Brewin 2011); that is, memories are not clearly differentiated.
Transcranial direct current stimulation over lateral occipital cortex during the
encoding of a memory leads to interference between memory representations,
presumably because of coactivation and less differentiation between those rep-
resentations (Koolschijn et al. 2019). Hence, such stimulation might poten-
tially be used as a system to model aspects of intrusive thoughts in PTSD.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Intrusive Thinking in Vivo and in the Context of Brain Mechanisms 93
thinking. For example, a stress response associated with traumatic events (e.g.,
increased cortisol and neurotoxicity of hippocampal cells) could cause reduc-
tion in hippocampal volume or shape. Moreover, examining brain anatomy
and resting-state connectivity may not be ideal for studying intrusive thought,
because both are relatively static, whereas intrusive thoughts, by nature, are
time limited and dynamic.
Other approaches characterize individuals according to their level and/or
controllability of intrusive thought to determine how these factors might af-
fect brain activation. For example, during suppression of recently experienced
items, individuals with a higher degree of self-reported difficulty in removing
current thoughts from consciousness had higher levels of activation in Broca
area; this presumably represents an inclination toward inner speech (Banich
et al. 2015). Another recent and potentially profitable approach is to use mag-
netic resonance spectroscopy (MRS) to examine potential neurochemical
mechanisms that may enable certain individuals to block memory retrieval.
For example, individuals with higher levels of GABA in the hippocampus, as
assessed by MRS, have a greater ability to suppress memory retrieval (e.g.,
Schmitz et al. 2017). The disadvantage of using MRS methods is that they are
quite time consuming (e.g., 25 minutes). In addition, only a few brain regions
can be interrogated during a scanning session, and the brain region interro-
gated is much larger (e.g., at least 3–4 times greater) relative to functional
neuroimaging methods.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
94 M. T. Banich
One important limitation of many of the methods described above is that they
rely on self-report to verify an intrusive memory or control over thoughts, and
thus do not provide insight into the nature of the representation of that memory.
Neurally based measures have been used to try to address this issue.
Autonomic Measures
Neuroimaging Approaches
Although much work on intrusive thinking has focused on the retrieval of in-
formation from long-term memory, by nature, intrusive thoughts involve ac-
cess to and active representation in working memory (for further discussion,
see Visser et al., this volume). Understanding whether a thought is currently
in the focus of attention in working memory is an important issue. Initial work
suggests that brain-imaging techniques can be applied quite fruitfully to verify
that individuals are indeed experiencing specific thoughts and/or manipulat-
ing them. In one study (Banich et al. 2015), individuals were shown a picture
or heard a brief snippet of a familiar tune (e.g., “Happy Birthday”) for four
seconds; immediately afterward they then had to manipulate the item in one of
four manners: maintain it, replace it with a different image/tune, specifically
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Intrusive Thinking in Vivo and in the Context of Brain Mechanisms 95
suppress the item, or clear their mind of all thought. Providing at least some ev-
idence that participants were indeed manipulating their thoughts as instructed,
a significant reduction of activity in primary sensory areas (e.g., visual cortex,
auditory cortex) was observed averaged across all trials for the two conditions
in which a thought needed to be cleared and removed (suppress the item or
clear their mind of all thought), compared to when there was an active thought
(maintain or replace). In addition, results indicated that while some neural
mechanisms were commonly engaged across these operations (e.g., across re-
placing, suppressing, or clearing an item as compared to maintaining it), there
are also specific neural mechanisms that differentiate the suppression of an
item from clearing one’s mind of all thought (Banich et al. 2015). Thus, neuro-
imaging may provide a neural marker and confirm specific mental operations
performed on a given thought. This may lead us to differentiate the ways in
which thoughts can be removed from working memory.
In follow-up work, machine learning was incorporated to expand the ques-
tions and issues that can be examined (Kim et al., submitted). Specifically,
via a localizer task, a machine-learning classifier was used to distinguish spe-
cific categories of items (e.g., places, faces, fruit). These classifiers were then
used on a trial-by-trial basis to determine the degree to which removing the
thought was successful. If a thought is successfully removed, then the clas-
sifier fit should be poor. If the thought is maintained, then the classifier fit
should be high. This approach enables us to examine the nature of the mental
representation on a trial-by-trial basis and provides a means to determine the
level at which such representations are maintained and/or cleared via specific
category and subcategory classifiers (e.g., fruit: apples, grapes, pears). For ex-
ample, one can examine whether clearing the thought of an apple generalizes
to other fruit (e.g., grapes and pears). In addition, patterns of brain activation
can be examined as a function of classifier fit to determine which brain re-
gions are highly active when the classifier fit is low (indicative of clearing the
thought), compared to when the classifier fit is high (indicative of the thought
remaining). This, in turn, may provide insight into which brain regions are
involved in exerting control over thoughts. While this study did not explicitly
track intrusions, such methods could be extended to track the content of intru-
sive thoughts on a real-time basis by identifying multivariate neural patterns
of distinct forms of intrusive content and evaluating the degree to which these
patterns manifest on a moment-to-moment basis.
In summary, there are a variety of interesting new directions that might be
used to provide novel insights into the nature of intrusive thought. These range
from adapting and using paradigms from other areas of cognitive and affec-
tive psychology to investigate intrusive thinking, to new approaches and ap-
plications from brain-imaging methods that might be used to verify or predict
intrusive thought.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
7
Conceptualizing Intrusive
Thinking at the Level of
Psychological Mechanisms
Marie-Hélène Monfils and David M. Buss
Abstract
Introduction
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
98 M.-H. Monfils and D. M. Buss
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Psychological Mechanisms 99
Though there is evidence that intrusive thoughts may be born from the recall
of previous experiences, this is not always the case. Intrusions can sometimes
consist of memories and other times not. There are instances in which the in-
trusion is autobiographical, but it is uncertain if that is always the case.
Many definitions of intrusive thoughts imply that they are negative in valence
(Clark 2005); however, Gregory et al. (2010) propose that intrusions may actu-
ally present as highly positive in individuals experiencing a hypomanic state.
These could be similar in content to thoughts related to delusions of grandeur.
Other situations suggest that positive intrusions exist, including in nonclinical
samples. One example is that of infatuation in which an individual experiences
intrusive thoughts about a loved one, and many such intrusions carry a posi-
tive valence (e.g., fantasies of union or sexual consummation). Even situations
that appear to have a negative connotation on the surface could carry positive
valence for the individual experiencing the intrusion. For example, imagining
the suffering of an enemy could be quite positive in a scenario of homicidal
ideation.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
100 M.-H. Monfils and D. M. Buss
The incidence of intrusions is quite high across multiple mental health disor-
ders, where they are known to occur in individuals with obsessive-compulsive
disorder (OCD), generalized anxiety disorder (GAD), PTSD, body dysmor-
phic disorder, eating disorders, depression, bipolar disorder, and others. The
manifestation of intrusions appears to be partly affected by specific diagnoses.
For example, an individual with OCD who engages in extreme handwashing
might experience germ intrusions, whereas an individual with body dysmor-
phic disorder might get intrusions related to food items. For a more extensive
discussion of intrusions in clinical populations, see Schlagenhauf et al. and
Visser et al. (this volume).
Interestingly, intrusions appear to be dissociable from other characteristics
that might be more specific to only one or two disorders. Whereas obsessions
are thought to be characteristic of OCD, worry is a central feature of GAD
(although not exclusive to it), and negative thoughts and rumination may typi-
cally be present in individuals with depression; intrusions are often present in
all of these conditions. Let us now compare and contrast intrusions with worry,
rumination, obsessions, and negative thoughts.
Defined as “a chain of thoughts and images, negatively affect laden and rela-
tively uncontrollable” (Borkovec et al. 1983), worry is a central feature in GAD,
but it also occurs with high incidence in nonclinical individuals. In definition
and in practice, worry and intrusive thinking are quite similar: They both inter-
rupt ongoing thoughts and activities, and they can both present as thoughts or
images, although worry occurs more frequently as verbal and intrusions more
frequently as images (Clark 2005). Intrusions are thought to be less voluntary
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Psychological Mechanisms 101
than worry; that is, worry can be brought on volitionally, whereas intrusions
are by definition involuntary and disruptive. Another distinctive feature that
disambiguates intrusions from worry is that intrusions are generally discrete
and brief, whereas worry need not be.
Obsessions and intrusive thoughts are very similar, where the former appears to
be an extreme version of the latter. Another characteristic that helps dissociate the
two is that intrusions may sometimes be irrelevant to the self, whereas obsessions
are relevant. Obsessions may often prompt behaviors such as compulsions that
are intended to diminish the associated thoughts and manifest as OCD.
Intrusions can be dissociated from general negative thoughts in that the former
is more likely to be irrational, whereas negative thoughts are more likely to be
rational. In this context, rational refers to thoughts that are not at odds with the
present context. An individual might be experiencing negative thoughts about
their promotion prospect during a recession, for example. If, during a positive
economy and after receiving a positive evaluation, they jarringly internally
hear the words “you’re about to get fired” just prior to giving an important
presentation, their experience was an intrusion. Intrusive thoughts are more
disruptive of day-to-day activity than general negative thoughts. Negative
thoughts are a core characteristic of individuals with depression, and generally
manifest as “thoughts” or in a verbal way rather than images. Intrusions may
present either as verbal or as images, most commonly the latter. Unlike other
forms of negative “processing,” intrusions seem to be relatively common in
nonclinical populations.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
102 M.-H. Monfils and D. M. Buss
Little is known about the etiology of intrusive thoughts. Different theories have
been proposed, but as yet we do not have a practical understanding of intru-
sions’ origins. Here, we briefly review various theories on intrusions, examine
their manifestation in nonclinical samples, and discuss the parallels between
intrusions and memory retrieval.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Psychological Mechanisms 103
What?
Why? How?
When?
Figure 7.1 Identifying why intrusions arise may help guide how we can best treat them.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
104 M.-H. Monfils and D. M. Buss
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Psychological Mechanisms 105
Individuals must constantly make decisions over choices and prioritize task
importance. It is conceivable that intrusions act as a means to emphasize
what should be worked through as soon as possible, in line with Salkovskis,
Rachman, and Klinger. The fact that intrusions appear suddenly, are brief, and
are often disturbing may emphasize the urgency of solving a potential prob-
lem. It is also conceivable that intrusions themselves present as a mechanism
to process information, in line with Horowitz. Here, we begin by examining,
in nonclinical populations, possible feelings that may be associated with intru-
sive thoughts, and we address their potential adaptive mechanism (intrusion
content). Thereafter, we approach the possibility that intrusive thoughts in and
of themselves are a mechanism that enables working through unaddressed but
identifiable problems by viewing intrusions through the lens of memory (intru-
sion process).
Though they are not necessarily centered around a disorder, intrusions in non-
clinical samples appear to occur along common themes across individuals. We
contend that these themes may provide insight into the potential adaptive na-
ture of intrusions, in line with the adaptationist framework proposed above.
The following list is not exhaustive, but provides a starting point to examine
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
106 M.-H. Monfils and D. M. Buss
different forms of emotions that can be at the source of intrusions. Their com-
mon characteristic is that they might serve an adaptive purpose; however, de-
pending on the context during which they occur, intrusions can also reflect an
adaptation gone awry.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Psychological Mechanisms 107
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
108 M.-H. Monfils and D. M. Buss
This illustrates that intrusive thinking is not solely a design feature of the
so-called “negative emotions.” It is likely an evolved design feature of many
emotions, including fear, rage, jealousy, shame, and guilt, as well as more
positively valenced emotions such as love and sexual arousal. Nor is it al-
ways dysfunctional. Intrusive thinking is often a key design feature, motivat-
ing attention to pressing adaptive problems while postponing effort allocated
to less important ones. Error management analysis highlights the difficulty of
distinguishing functional from dysfunctional outcomes in any specific case,
rendering the theoretical analysis of intrusive thinking more complicated than
previously considered. Another example that generally involves positive in-
trusions pertains to the pursuit of goals or aspirations. It is important to note
that while some emotional contexts are conducive to either positive or nega-
tive intrusions, others are likely to be more complex, as in the case of grief.
Intrusions can often be autobiographical, that is, they relate to an individual’s
firsthand experience. Cast in this light, intrusions can actually be interpreted as
a memory retrieval in some instances. We briefly consider intrusions occurring
in the context of goals or aspirations and grief, and then further examine intru-
sions as memory retrieval to extract their potential adaptive process.
Grief. Grief is almost always triggered by the loss of a key social partner—a
close friend, a romantic partner, or a family member. Research on intrusions dur-
ing grief is limited but suggests the presence of both positive and negative intru-
sions in individuals experiencing the loss of a loved one (Boelen and Huntjens
2008). In the positive realm, mourners may experience intrusive memories of
the loved one that died or fantasy reenactment. Through the lens of memory
(discussed below), the purpose of grief intrusions could be that of strengthening
a neurobiological trace, to keep the memory alive. Negative intrusions of grief
can include memories of the death event or negative images or thoughts about the
future. Early during grief, the intrusions, both positive and negative, may be help-
ful to the individual who experienced a loss. However, if persistent and enduring,
they could interfere with a person’s ability to move forward.
Evolutionary scholars have advanced two competing explanations of grief.
One is that grief is an unfortunate nonadaptive by-product of love and attach-
ment (Archer 2003), both of which are profoundly important adaptations in
the evolved social suite of humans (Christakis 2019). The second is that grief
serves several adaptive functions, such as identifying actions that might have
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Psychological Mechanisms 109
led to the loss, motivating actions to prevent future losses and signaling to sig-
nificant others (friends, family, mates) the need for help due to the loss (Nesse
2005). Other possible functions include signaling to others that you are a loyal
coalitional ally and ruminating about the implications of the loss for replacing
the lost one with an alternative mate or coalitional ally. Which of these compet-
ing hypotheses, or which combination, will bear fruit rests with future empiri-
cal research on individuals who experience loss and grief.
There are useful parallels to be drawn between intrusions and other forms of
memory retrieval. Memory retrieval can be broadly defined as recalling a prior
experience, either following the presentation of an external or internal cue or
through volitional control. Conceptualizing intrusions as memory retrieval ap-
pears very much in line with Horowitz’s definition of intrusive thoughts and
enables approaching the concept with an adaptive mechanistic view.
From this perspective, we can think of intrusions as potentially serving the
adaptive functions described below. We can also conceive of intrusions as pro-
viding an opportunistic window or intervention. The latter can best be under-
stood through the process of reconsolidation and memory updating.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
110 M.-H. Monfils and D. M. Buss
Summary
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Psychological Mechanisms 111
intrusion has not per se caused any harm. The person’s reaction to the intru-
sion, however, can vary. One individual could simply think: “Whoa! That was
crazy! Of course, I would never do that.” At the same time, this person might
experience a sense of comical relief: “Boy, if I had jumped, I wouldn’t have
to sit through all of those blankety-blank-blank meetings scheduled today.”
Another person might process the intrusion differently. For instance, they
might think: “Why am I having these images? Should I jump? I am worthless.
I don’t want to die. Or maybe I do want to die. I don’t know what to do. I’m
worried about what I might do.”
In the first response scenario, the person may not be bothered further (or
at all) by the intrusion. In the second case, a person might perseverate on the
experience or engage in behaviors to try to minimize the impact of the intru-
sion. This could potentially result in worry, rumination, and/or obsessions and
associated compulsions.
In a sense, intrusions themselves may not be as distressing as what we make
of them, and what we make of them is likely to be largely influenced by our
state of mind (including, in clinical manifestations, the underlying pathology).
In approaching treatment for individuals who experience intrusions, it is im-
portant to consider their possible adaptive nature. In doing so, it may be helpful
to identify the intrusions’ possible underlying content as well as the psycho-
logical process that a person’s brain has determined should be engaged via the
intrusive thought. Ultimately, identifying the possible “why” of intrusions may
help guide “how” we can best treat them (Figure 7.1).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
8
Abstract
Various psychopathological symptoms share characteristics of intrusive thinking. Intru-
sive thoughts are part of the diagnostic criteria for posttraumatic stress disorder and ob-
sessive-compulsive disorder but are also relevant in other psychiatric conditions, such
as drug craving in addiction or rumination in depressive disorders. Intrusive thoughts
must be differentiated from thought insertion observed in schizophrenia and related
psychotic disorders. This chapter reviews the typical characteristics and content of in-
trusive thinking in the context of different psychiatric conditions and outlines current
theories regarding the mechanisms of intrusive thinking.
Introduction
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
114 F. Schlagenhauf, A. Heinz, and M. Voss
One prominent and required diagnostic criterion for PTSD is the presence of
intrusion symptoms: a traumatic event is persistently reexperienced. According
to DSM-5, Criterion B (American Psychiatric Association 2013), these intru-
sion symptoms encompass the following characteristics:
• Recurrent, involuntary, and intrusive memories of the traumatic event(s)
• Traumatic nightmares
• Dissociative reactions (e.g., flashbacks)
• Intense distress after being exposed to traumatic reminders
• Heightened physiologic reaction to trauma-related stimuli
To meet the full diagnostic criteria, a person must have been exposed to a life-
threatening event and thereafter has to avoid trauma-related stimuli. Further,
the person exhibits negative alterations in cognition and mood, which began
or worsened after the traumatic event, and displays alterations in arousal and
reactivity, such as hyperarousal.
The clinical characteristic of intrusive memory is that it “springs to mind
unbidden—that is, against the person’s will” (Visser et al. 2018). Such intru-
sive memories are forms of episodic memories of actually experienced auto-
biographical events, which are retrieved involuntarily. In its extreme form, the
person intensely and vividly relives the traumatic event in the present. Such
flashbacks involve the retrieval of detailed sensory features and are highly
emotional. Typically, fragments and several distinct moments of the trauma
are recalled, the so-called “hot spots,” in a predominantly visual form (Visser
et al. 2018).
Disturbance in memory seems to be the prominent feature of PTSD. It is
widely agreed that multiple memory systems exist and that these rely, in part,
on distinct neurobiological substrates (Henke 2010). “Declarative” memories
are events and facts, which we can explicitly remember, and seem to depend on
medial temporal lobe structures. “Nondeclarative” memories are implicit and
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Current Psychiatric Perspectives on Intrusive Thinking 115
Obsessive-Compulsive Disorder
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
116 F. Schlagenhauf, A. Heinz, and M. Voss
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Current Psychiatric Perspectives on Intrusive Thinking 117
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
118 F. Schlagenhauf, A. Heinz, and M. Voss
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Current Psychiatric Perspectives on Intrusive Thinking 119
(who later relapsed) from abstainers and controls only when individual alco-
hol expectancies were considered. This suggests that habitual behavior does
not generally increase in addicted patients during choice behavior tests for
non-drug-related rewards in a laboratory context; addiction-related habits ap-
pear to be triggered by specific cues and contexts in conjunction with previ-
ous experiences. In line with this hypothesis, a patient suffering from OCD,
pathological gambling, and drug addiction described constant urges to perform
habits compulsively related to his obsessions, while craving for gambling and
drug intake was triggered only during certain time periods by specific drug or
gambling-related stimuli (Schoofs and Heinz 2013). In light of these findings,
compulsions in OCD appear to differ significantly from “compulsive” urges to
consume drugs of abuse, warranting further phenomenological and neurobio-
logical specifications (Heinz 2017).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
120 F. Schlagenhauf, A. Heinz, and M. Voss
expressed by Feinberg 1978 and Ito 2008) and that every thought is preceded
by an intention to think this thought. The actual thought occurring in one’s
stream of consciousness is then compared with the intention to think. When
these two processes match, a feeling of authorship results; when they do not,
attribution to another agent may occur.
While this offers an appealing framework to explain passivity phenomena,
such as delusions of control or hallucinations, several problems arise when the
comparator model is used to explain thought insertion (for a detailed critique,
see Vosgerau and Synofzik 2010). One problem is that the account does not
distinguish thinking (as a process to generate thoughts) from thoughts (as a
result of thinking), making it difficult to pinpoint the difference between “in-
fluenced” thinking and “inserted” thoughts. Furthermore, it remains unclear
what an intention to think a specific thought could be and how it can be dis-
tinguished from the actual thoughts; that is, why the intention to think is not
naturally conceived of as the thought itself. Indeed, if every thought were to
be preceded by an intention to think, we would run into an infinite regress: for
each thought, we need a thought to get the process started, which in turn pre-
supposes another thought, and so on.
A more recent attempt to conceptualize the phenomenon of thought inser-
tion treats inserted thoughts as sensory events rather than motor processes
(Sterzer et al. 2016). Building on detailed phenomenological descriptions
from the early Heidelberg school in the first half of the twentieth century,
which described thoughts that “become sensory” as being experienced as in-
serted, more recent studies explain thought insertion within the framework of
predictive coding and Bayesian inference (Sterzer et al. 2016). Here, thought
insertion is viewed not as the failure of introspection in a comparator process
(Campbell 1999) but rather as the failure (or imprecision) of prior beliefs
in a Bayesian inference process thought to be at the core of thought inser-
tion (Sterzer et al. 2016). In analogy with aberrant salience attribution to
external events, which could lead to the emergence of delusional mood and
fixed beliefs (Heinz 2002; Kapur 2003; Heinz and Schlagenhauf 2010), in-
ternal events (e.g., verbalized thoughts) may also be experienced as overly
salient, and therefore unusual, as well as surprising due to a lack of context
and unusual structure (with a possible link to formal thought disorder). The
individual’s attempt to explain the aberrant salience and unusual character of
such verbalized thoughts could result in their interpretation as being exter-
nally caused. Nonetheless, unintended or semantically inappropriate verbal-
izations can also be expressed in aphasia (e.g., due to a stroke) but are not
accompanied by reports of “alien” involvement (Heinz 2017). Therefore, ad-
ditional steps may be required to convince a person that a thought is “alien”
and thus “must” be inserted by an external agent. In this context, it has been
suggested that low precision of prior beliefs and/or increased sensory preci-
sion, both inaccessible to introspection, may render some thoughts so unpre-
dictable that they are experienced as inserted (Heinz et al. 2019). Whereas
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Current Psychiatric Perspectives on Intrusive Thinking 121
beliefs and desires may have a role in prior beliefs regarding our own
thoughts (Stephens and Graham 2000), what makes a thought feel alien is
not that they are simply unwanted or “immoral” (in the sense of higher-level
introspection), but rather that patients directly notice or perceive a thought
to be “alien,” thus pointing to unconscious mechanisms causing this experi-
ence. Such mechanisms may be usefully described by a Bayesian account of
information processing in the central nervous system (Adams et al. 2013):
during psychotic episodes, if prior beliefs are indeed imprecise in compari-
son to sensory input-driven posteriors, frequent prediction errors are made,
which may trigger phasic dopamine release and cause salience to be attrib-
uted to otherwise irrelevant cues, including verbalized thoughts (Sterzer et
al. 2016; Heinz et al. 2019).
If these considerations are correct, they may help to explain the difference
between “alien” thought insertion and other forms of unwanted or intrusive
thought content. In psychosis, impaired precision of prior beliefs may affect
the whole experience of the world and self, which are experienced as unusual,
alien, and often threatening. Thus, thought insertion in psychosis goes beyond
the experiencing of unusual verbalizations (as in aphasia), unwanted cravings
(in addiction), self-centered concerns (in major depression), intrusive memo-
ries (in PTSD), or obsessive thoughts (in OCD). The entire relationship be-
tween the individual and the real world is affected: that which was well known
for a long time suddenly carries hidden meanings, harmless situations are im-
bued with a sense of danger, and longtime friends and family members become
deeply alienated and may no longer be trusted. Since Bayesian accounts are
supposed to reflect general functions of the central nervous system, such com-
putational frameworks will have to explain how more restricted alterations in
information processing, particularly with respect to verbalized thoughts, differ
from the more fundamental alterations experienced in psychotic states.
Open Issues
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
122 F. Schlagenhauf, A. Heinz, and M. Voss
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
9
Neuropsychological
Mechanisms of
Intrusive Thinking
Renée M. Visser, Michael C. Anderson, Adam Aron,
Marie T. Banich, Kathleen T. Brady, Quentin J. M. Huys,
Marie-Hélène Monfils, Daniela Schiller, Florian Schlagenhauf,
Jonathan W. Schooler, and Trevor W. Robbins
Abstract
A classic definition of intrusive thinking is “any distinct, identifiable cognitive event
that is unwanted, unintended, and recurrent. It interrupts the flow of thought, inter-
feres in task performance, is associated with negative affect, and is difficult to control”
(Clark 2005:4). While easy to understand and applicable to many cases, this definition
does not seem to encompass the entire spectrum of intrusions. For example, intrusive
thoughts may not always be experienced as unpleasant or unwanted, and may in some
situations even be adaptive. This chapter revisits the definition of intrusive thinking, by
systematically considering all the circumstances in which intrusions might occur, their
manifestations across health and disorders, and develops an alternative, more inclusive
definition of intrusions as being “interruptive, salient, experienced mental events.” It
proposes that clinical intrusive thinking differs from its nonclinical form with regard to
frequency, intensity, and maladaptive reappraisal. Further, it discusses the neurocogni-
tive processes underlying intrusive thinking and its control, including memory pro-
cesses involved in action control, working memory and long-term memory encoding,
retrieval, and suppression. As part of this, current methodologies used to study intrusive
thinking are evaluated and areas are highlighted where more research and/or technical
innovation is needed. It concludes with a discussion of the theoretical, therapeutic, and
sociocultural implications of intrusive thinking and its control.
Group photos (top left to bottom right) Renée Visser, Michael Anderson, Marie-
Hélène Monfils, Trevor Robbins, Daniela Schiller, Marie Banich, Adam Aron,
Kathleen Brady, Jonathan Schooler, Florian Schlagenhauf, Quentin Huys, Michael
Anderson, Daniela Schiller, Jonathan Schooler, Kathleen Brady, Quentin Huys, Florian
Schlagenhauf, Marie-Hélène Monfils, Trevor Robbins, Renée Visser, Marie Banich
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
126 R. M. Visser et al.
Introduction
While reading this chapter, you might be sitting in a coffee shop somewhere
or traveling on a train. The sunlight comes in through the window and you
envision how warm the planet might get. You are having a meeting in a few
hours and various scenarios of it are running through your head. Perhaps these
few examples have triggered some of your own thoughts, as you are no longer
reading this article: your eyes are just glazing over these words. If this just
happened and you rejoin us a few sentences below, you have just experienced
an intrusive thought.
We all seem to have an intuitive understanding of what an intrusion is.
Yet, attempts to define it immediately sets off debate. According to the one
textbook that has been written on this is, the classic definition of an intru-
sive thought is “any distinct, identifiable cognitive event that is unwanted,
unintended, and recurrent. It interrupts the flow of thought, interferes in task
performance, is associated with negative affect, and is difficult to control”
(Clark 2005:4). Central to this definition is the assumption that an intru-
sive thought always constitutes an unpleasant experience which negatively
impacts functioning. Here, we revisit the definition of intrusive thinking,
by considering all the circumstances in which intrusions might occur, their
manifestations across health and disorders, and their neurocognitive basis.
We start with a rather narrow, presumably more commonly accepted defini-
tion of intrusions being conscious, involuntary, unwanted thoughts, and ar-
rive at an alternative, more inclusive definition of intrusions as interruptive,
salient, experienced mental events. We discuss current methodologies used
to study intrusive thinking, highlighting existing strengths as well as areas
where more research and/or technical innovation are needed. We conclude
with a discussion of the theoretical, therapeutic, and societal implications of
intrusive thinking and its control.
Although intrusive thoughts are often associated with mental health disor-
ders, they also occur to healthy individuals in everyday life (Purdon and Clark
1993; Berntsen 1996). Identifying broad circumstances under which these
thoughts occur to people in general, irrespective of mental health status, may
highlight their adaptive functions in healthy individuals and illuminate the
processes that generate such intrusions. Moreover, considering the motiva-
tions that people may have for controlling intrusions in daily life can shed
light on important psychological and social functions that the capacity for
control helps to support.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 127
Although unintended thoughts can occur for any content, not all such thoughts
are intrusive and bothersome. There are, however, certain contents that consis-
tently trigger intrusive thoughts in otherwise healthy people. Here we discuss
several examples and consider why this content is intrusive.
Incompletions. When people initiate a process and are then unable to com-
plete it, due to an interruption or some other impasse, thoughts related to that
incompletion tend to recur until the process is completed (Horowitz 1975).
This hypothesis is reflected in the classical Zeigarnik effect proposed in the
early twentieth century by Gestalt psychologists, who posited that people had
superior memory for interrupted processes (Zeigarnik 1938). Incompletions
can include both physical and mental tasks, as long as there is an unresolved
problem. An interesting possibility is that the tendency for incompletions to
precipitate intrusive thoughts may be amplified by the salience or emotional
intensity of the incomplete process (Horowitz 1975). Intrusive thoughts related
to the interrupted process (i.e., seeking comprehension, solving a problem)
may persist until the situation is understood. This suggests that emotionally
intense events may not intrude merely due to salience or encoding strength,
but because they are accompanied by a compelling desire to comprehend them.
Relatedly, when people reach a significant impasse in solving a difficult
problem, they continue to “work on the problem” in the background. For ex-
ample, research on creativity and insight problem solving suggests that when
people allow a period of incubation, insights may emerge spontaneously (the
“aha” phenomenon), as though a process has occurred in the background (Gable
et al. 2019). The commonly experienced “tip of the tongue” phenomenon in
memory retrieval (described initially by W. James and S. Freud) is another
example of when a temporarily forgotten item finds its way back into con-
sciousness unpredictably. The content of people’s mind wandering (i.e., when
thoughts distract us from a task at hand) often includes unresolved problems
salient to the individual, whether emotionally significant or not (Smallwood
and Schooler 2006; Klinger and Cox 2011).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
128 R. M. Visser et al.
Uncertain Events. When people are uncertain about a past or future event,
this can promote intrusive thoughts for several reasons (Grupe and Nitschke
2013). For example, if someone is uncertain about whether they have already
performed an action (e.g., locking the front door or taking one’s pills), associ-
ated uncertainty may precipitate worry. Uncertainty about future outcomes
can also result in persistently intruding thoughts, which in turn may prepare
people for different outcomes, thus enabling them to be ready to respond
appropriately.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 129
2013, 2015; Moeck et al. 2018). Research suggests that whether an ear worm
develops is related to how often the song has recurred in recent experience.
The previous discussion reviews content that is especially prone to generate in-
trusive thoughts, but it is worth separately considering the conditions that gen-
erate intrusions. In general, the likelihood of intrusions occurring in a given
moment is related to both the presence of retrieval cues as well as the availability
of control resources to resist unwanted intrusions. Here we discuss the critical
role of cue-driven retrieval, matching physiological and mood state, monitoring
processes that may cue thoughts, and the availability of inhibitory control.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
130 R. M. Visser et al.
Matching Mood and Physiological State. Sometimes the cues that trig-
ger retrieval are not specific stimuli in the world or concepts in memory, but
rather are broad psychological or physiological states (Berntsen 1996). It is
well established that information and experiences are often stored in memory
in association with a representation of the context in which they took place
(for a review, see Anderson and Hanslmayr 2014). This can include not only
the spatiotemporal context (i.e., the environment), but also the mood, drug, or
arousal state present at encoding. Later on, the chance of retrieving the content
is higher when a person’s state resembles the one at encoding. For example,
experiences encoded in a sad or angry mood are more likely to be recalled at a
later time, when a person is once again in a sad or angry mood than when they
are in a happy mood.
Desirability of Control
In healthy individuals, when intrusive thoughts occur, they are usually per-
ceived as unwanted, at least at that moment. As a result, people often try to
exclude them from awareness in an effort to regain control over thoughts and
emotions. When successful, such efforts enable a person to put unwelcome
thoughts out of mind, thus diminishing their accessibility in memory and re-
ducing their tendency to return. Such attempts to facilitate the forgetting pro-
cess often serve important behavioral, emotional, and social functions. Here,
we consider several contexts in everyday life in which people are motivated to
forget thoughts for a functional reason.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 131
Regulating Pain. Thoughts about pain, what it feels like, how uncomfortable
it is, and worries about what it may mean often intrude persistently, even after
the pain has resided. We consider people who can manage intrusive thoughts
about their pain to be resilient and able to cope. In contrast, those who are over-
whelmed by pain, who allow their discomforts to blossom into catastrophic
thinking, not only pay a price for this distraction, their suffering becomes mag-
nified and extended (Edwards et al. 2006). When confronted with a painful
stimulus (e.g., the cold pressor task), people with a moderate history of adver-
sity tolerate the pain longer and show less pain catastrophizing (the label given
to excessive thoughts about the pain), relative to people who have no history
of trauma (Seery et al. 2013). It appears, therefore, that in the face of physical
discomfort, resilience emanates from the capacity to control intrusive thoughts
about the painful stimulus, which not only reduces suffering but increases the
ability to pursue other goals.
Regulating Affect. When unwelcome thoughts intrude, they often evoke un-
welcome changes in emotional state. For example, being reminded of an ar-
gument may trigger anger; images of an upcoming doctor’s visit may trigger
fear or anxiety; or seeing the same car that your ex-partner used to drive may
evoke sadness. As a result, intruding thoughts trigger mechanisms that regu-
late emotion to return a person to a neutral or positive state. Although these
endogenously triggered emotions are common and contribute substantially to
psychological disorders, they have been neglected in research on emotion reg-
ulation, which has focused on regulating affect triggered by external emotion-
eliciting stimuli. Understanding how intrusive thoughts can be downregulated
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
132 R. M. Visser et al.
in memory may provide critical insights into how people regulate the emotions
that they elicit, contributing to successful emotional homeostasis (Engen and
Anderson 2018).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 133
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
134 R. M. Visser et al.
Unintended. Individuals frequently report mind wandering and are often not
even aware that they are engaged in it, until they are caught, as in the expe-
rience sampling probe (Schooler et al. 2011). Nevertheless, individuals may
sometimes deliberately abandon a task in favor of other thoughts (Seli et al.
2015). The lack of awareness that one is mind wandering (i.e., lack of meta-
awareness; Schooler 2002) contributes to more disruptive mind wandering
episodes (Schooler et al. 2011), which are neurologically distinct from on-task
thinking (Christoff et al. 2009). Notably, lack of meta-awareness has been as-
sociated with mind wandering about unwanted thoughts (Baird et al. 2013a).
This suggests that intrusive thoughts may similarly slip below the radar of
meta-awareness (Takarangi et al. 2014).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 135
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
136 R. M. Visser et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 137
with the mental act. The thought is intrusive because of its obsessive nature.
She cannot suppress the thought. Moreover, she is compelled to think about
her obsession. Compulsivity is a dysfunction of sense of agency: she is forced
to think about the intrusion, contrary to her willpower. The thought is intru-
sive because of its compulsive nature. Gradually the thought becomes more
present and repetitive; it loses its original meaning, but becomes intrusive
because of its duration and repetition. The thought has become a full-blown
obsession. Obsessions are answered with compulsions (e.g., obsessing about
germs could lead to compulsive handwashing). Though initially successful
in reducing anxiety, gradually they become intrusive since the acts have to
be performed compulsively. Note that both obsessions and compulsions are
intrusive, and both have an obsessional and compulsive quality. Eventually,
the anticipatory power of the intrusion is so overwhelming that reality test-
ing gets disturbed. She does not know anymore whether she has or has not
strangled her baby. Thoughts may take on a delusion-like character, with psy-
chotic features.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
138 R. M. Visser et al.
Mood Disorders
Clinically, intrusive thoughts are neither a core nor a necessary feature of mood
disorders. However, they appear regularly in a manner that has important bear-
ing on its treatment, severity, and outcome, specifically in the form of rumina-
tion, suicidal thoughts, negative automatic thoughts and “flight of ideas.”
Rumination is a frequently observed cognitive feature of depression, in-
volving the repetitive and persistent focusing on the causes of the current
state of distress and its likely consequences. Depressive rumination is typi-
cally centered around personal shortcomings, faults, failings, and mistakes
(Treynor et al. 2003). Rumination has intrusive features: it is highly interrup-
tive and distracts individuals from engaging in other tasks. While individuals
often state they do not want to ruminate, they also hold metacognitive beliefs
that it is important to do so (Borkovec and Roemer 1995). Rumination is asso-
ciated with the onset of depression and, when combined with negative cogni-
tive styles, predicts the duration of depressive symptoms (Nolen-Hoeksema et
al. 2008). The extent to which it is hence desired and indeed under volitional
control is unclear.
Individuals suffering from suicidal thoughts report at times imagery related
to potential ways of committing suicide (Holmes et al. 2007). This imagery can
be experienced as unwanted, intrusive, and interruptive.
Negative automatic thoughts are regarded as a central feature of depres-
sion in cognitive models of depression (Beck 1976). In this view, events that
are related in some form to core negative beliefs or schemata can trigger fast
interpretations or “automatic thoughts,” which, due to their negative nature,
promote a depressed, negative mood. These automatic thoughts appear very
rapidly and can profoundly influence behavior. However, as these automatic
thoughts closely relate to core beliefs, they tend not to be experienced as intru-
sive or unwanted.
Individuals with manic episodes of bipolar disorder can display a “flight
of ideas,” whereby they exhibit a rapid sequence of unrelated thoughts. This
state is often experienced in the context of mania and tends to be positively
experienced as a phase of heightened creativity. In mixed states, however, it
can also be perceived negatively and exhibit loss of control of one’s thinking.
Hence, although intrusive thoughts are not central to mood disorders, they do
feature prominently and in areas that are thought to be closely related to the
core mechanisms of the illnesses.
Anxiety
Two subtypes of anxiety have been distinguished: anxious arousal (e.g., panic)
and anxious apprehension (e.g., worry). In cases of anxious arousal, the in-
dividual has fearful reactions and thoughts that lead to somatic symptoms
and/or somatic symptoms are interpreted in a fearful manner. For example,
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 139
in simple phobias, an individual will see a phobic object (e.g., a spider) and
have intrusive thoughts regarding that object (e.g., “The spider is about to
crawl on me and bite me”) that are often accompanied by somatic symptoms
(e.g., sweating, increased heart rate). In cases of panic disorder, the individual
has somatic symptoms, such as increased heart rate and dizziness, associated
with intrusive thoughts (e.g., “I am having a heart attack and am going to
die”). This situation of a bodily state leading to intrusive thoughts is some-
what akin to SUDs, in which a somatic state (i.e., withdrawal) can lead to
intrusive thoughts of craving.
In contrast, in cases of anxious apprehension, intrusive thoughts tend to
be related to a future event: about the feared event itself, ways to avoid the
fear, or the discomfort/harm that might be associated with the event. In some
cases, intrusive thoughts can be more distressful when a person anticipates
the event, than during the event itself. This appears to be associated with
the fact that individuals with anxious apprehension have an intolerance of
uncertainty (e.g., not being 100% sure that the flight will not involve severe
turbulence).
Individuals that experience anxious arousal and/or anxious apprehension
may know that their fears are likely unrealistic or overblown. This knowledge,
occurring simultaneously with the experienced fear and intrusive thought,
can cause distress and lead to additional intrusive thoughts related to poor
self-evaluation similar to depressive rumination: “What is so wrong with me
that I cannot get on an airplane like everyone else? Why am I such a baby?”
However, intrusive thoughts associated with anxiety tend to be more future
oriented whereas intrusive thoughts associated with depression tend to be more
concerned with past events.
Psychosis
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
140 R. M. Visser et al.
for the person’s relation to their interpersonal and social surrounding. A special
case is delusion of control such as thought insertions, when (mostly verbal)
thoughts are experienced as being “inserted” by another agent into one’s head.
Inserted thoughts can be salient, interruptive to the train of thought, and un-
wanted, but are attributed to an outside agent rather than to oneself, and there-
fore cannot be experienced as ego-syntonic nor ego-dystonic.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 141
Salience. Something can only interrupt if it has gravitational pull; that is, if it
captures attention (see also Fedota and Stein, this volume). Such gravitational
pull may lie in different, not mutually exclusive, aspects (e.g., its valence,
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
142 R. M. Visser et al.
highly negative or positive), its vividness, its novelty, or its incongruence with
a current state.
Content and Shape. Intrusions can take different shapes: they can present as
verbal thoughts, slips of action, or mental images. Content varies in valence
as well as in time frame, ranging from the past, present, or future. We suggest
not including content or shape as a defining feature of intrusive mental events
as it might vary widely between different instantiations of intrusive thinking in
healthy and clinical states.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 143
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
144 R. M. Visser et al.
Experienced
Interruptive
Involuntary
Conscious
Unwanted
Salient
Symptom Related disorder Further specifier
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 145
assessed. As suggested above, situations that trigger intrusions all tap into
mechanisms related to the ability to control thoughts, actions, and mnemonic
processes. Most of these processes come into play at the time of (involuntary)
memory retrieval or when habits or compulsions interfere with goal-directed
behavior. Here, we elaborate on what is known about these mechanisms and
the methods to investigate them.
Action Control
Goal-directed actions are those that we acquire to bring about a change in the
world that accords with our basic desires. In this view, I perform an action, A,
because I want a certain outcome or goal, O, and believe that A is the way to
achieve O. Action A may start life under goal-directed control; however, with
extended training performance, it can become more automatic and invariant,
elicited by environmental stimuli, instead of to achieve specific consequences,
thus becoming habitual. Both actions and habits are acquired and represented
in parallel rather than serially. Therefore, it is possible for certain errors to
emerge as a consequence of two controllers attempting to control action si-
multaneously: suddenly switching from goal-directed to habitual control can,
for instance, result in the intrusive performance of an action that is adaptive
in another situation, but not in the present situation where it is unwanted (e.g.,
continually turning on the windscreen wipers rather than the blinkers in a
rental car). Conversely, intrusive goal-directed control can disrupt the smooth
operation of a well-learned habit. It is well known, for instance, that suddenly
exerting goal-directed control while playing a complex piece of music on the
piano can result in reduced capacity to perform the piece accurately. The same
holds for well-trained athletes (see Balleine, this volume).
Both actions and habits require top-down control to prevent excessive and
hence maladaptive impulsive behavior. Clues about neural systems support-
ing the regulation of intrusive thoughts come from the field of action control,
capitalizing on putative similarities between inhibition of motor responses and
thought control. In the simple Stop-Signal paradigm, for each trial, subjects
get ready to respond when a Go signal occurs; then, on a minority of trials,
they try to stop in response to a subsequent Stop signal (Lappin and Eriksen
1966; Verbruggen and Logan 2008). Anderson and Green (2001) have long
suggested that stopping a memory or thought from intruding is like stopping a
motor response. When the stopping process fails, an intrusion occurs. Several
studies have provided evidence consistent with this, although it is not yet cer-
tain whether there is a common process, common principle, or common cir-
cuitry (e.g., Morein-Zamir et al. 2010; Depue et al. 2016; Guo et al. 2018;
Castiglione et al. 2019). Assuming there is a commonality between the stop-
ping of a motor response and the termination of a memory or thought, it may be
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
146 R. M. Visser et al.
useful to adopt a distinction from the motor stopping literature between global
suppression and selective suppression (Aron 2011). Stopping a memory or
thought can be achieved by a global systemic suppression mechanism, which
suppresses all memories (Hulbert and Anderson 2018), or through a selective
stopping process in which the target is suppressed by retrieving an alterna-
tive memory (Hertel and Calcaterra 2005; Bergström et al. 2009; Benoit and
Anderson 2012).
Pavlovian-instrumental transfer (PIT) provides another experimental par-
adigm relevant to intrusive thinking and may provide evidence of intrusive
thinking in nonhuman animals as well as humans. In the case of positive PIT,
experimental animals (or humans; Freeman et al. 2014) are trained on two
actions to earn distinct reward outcomes: A1 → O1 and A2 → O2. They are
then exposed to two stimuli paired with the same outcomes but without the
opportunity to perform the actions: S1 – O1, S2 – O2. They are then allowed
to perform the two actions in extinction (without reward) but in the presence
of the two stimuli. In this situation, the animals will perform the two actions
at a low baseline rate. When one of the other stimuli is presented, however,
this leads to an interruption in the animals’ performance and causes them to
immediately shift performance to the action which, in training, earned the goal
or outcome predicted by the stimulus (specific PIT). General PIT occurs when
the stimulus reminds the subject instead of the valence of the goal and simi-
larly may enhance responding nonspecifically. Probably even more relevant
to intrusive thinking is negative or aversive PIT in which the occurrence of a
negative Pavlovian conditioned stimulus (associated, e.g., with painful electric
shock or loud white noise) enhances avoidance or withdrawal behavior or dis-
rupts appetitive behavior, as in conditioned suppression.
These findings suggest that the presentation of the stimulus during the task
might intrude on the animals’ ability to perform their continued action. In this
sense, the stimulus might be regarded as an intrusion, bringing to mind an
outcome. The occurrence of the interruptive stimulus reminds the animal of its
associated action, the performance of which provides a readout of that intru-
sion. PIT is not influenced by extinction of the predictive learning nor by the
devaluation of the instrumental outcome. In these cases, the stimuli continue to
drive the performance of a specific action, demonstrating continued stimulus-
mediated retrieval of the specific, now unwanted, outcome.
Working Memory
While much research relevant to intrusive thoughts, such as those that occur
in PTSD, has focused on long-term memory formation and retrieval (the topic
of the next sections), significantly less work has been done on the role that
working memory mechanisms may play in intrusive thoughts. Yet, by their
very nature, intrusive thoughts can only be intrusive to the degree that they can
gain access to working memory and hijack an individual’s attention. Working
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 147
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
148 R. M. Visser et al.
preceding set of four. On certain trials, the probe is one of the four items from
the prior trial, but not the current one. In such cases, individuals show reduc-
tions in performance compared to a less recent probe. This paradigm demon-
strates that time can be a marker for controlling what information is within the
current focus on attention. Future research might explore the degree to which
selection (either on the basis of perceptual, topical, or temporal characteris-
tics) is affected in individuals who experience intrusive thoughts. To the degree
that these control mechanisms are defective, they may not allow for intrusive
thoughts to be moved from the focus of attention, or manipulated or reinte-
grated in such a way as to reduce their saliency (e.g., via reappraisal).
In the third process, information must be removed at some point from work-
ing memory to allow more relevant information to be placed in this limited
buffer. Theoretically it has been argued that removal may occur via three po-
tential mechanisms: (a) passive decay, (b) being replaced by something else,
or (c) being actively removed (Lewis-Peacock et al. 2018). Thoughts may be
intrusive to the degree that they are particularly resistant to such processes.
These processes may be especially important for disorders of intrusive thoughts
which are more abstractly cognitive (i.e., not sensory based) and more repeti-
tive in nature. Moreover, in certain cases it may just be specific categories of
information that are specifically resistant to removal. For example, the degree
of rumination observed in depressed individuals is associated with difficulty
in removing negatively biased information from working memory (Joormann
and Gotlib 2008). This difficulty may occur because of an entrenched overrid-
ing schema for thought that focuses attention to negative material.
Long-Term Memory
Many intrusive thoughts concern contents that have been encoded into long-term
memory. This content includes not only personal autobiographical memories
of past events, but also other contents that clearly rely on declarative memory
but are not autobiographical memories per se (e.g., images of feared future
scenarios or memories of prior thoughts). As noted earlier in this chapter, these
intrusive thoughts are often elicited by retrieval cues in the environment or, in
some cases, cues that are internally generated by the person. While sometimes
positive or neutral, intrusions often have a negative tone. In their extreme form,
intrusive memories involve vivid, multisensory images from highly aversive
events, constituting the hallmark symptom of disorders such as PTSD. When
looking at the role of mnemonic processes underlying intrusive memories and
opportunities to intervene, there are different stages of memory that need to be
considered. A distinction can be made between the encoding of a memory and
its retrieval: within the context of encoding, processes of encoding suppression
can limit the impact of an event; within the context of retrieval, one can focus on
opportunities for memory modification as well as mechanisms for suppression.
Critically, intrusive thoughts involving content encoded into long-term memory
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 149
Memory Encoding
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
150 R. M. Visser et al.
in a list, the instruction has a large impact on later retention of the memory,
rendering to-be-forgotten items significantly less accessible, even for emo-
tionally unpleasant scenes (Anderson and Hanslmayr 2014). Such directed
forgetting effects arise not only on recall tests, but also on recognition and
implicit memory tasks, indicating that participants can successfully block
the encoding of recent events into long-term memory when motivated to do
so. This encoding suppression mechanism is likely one coping process that
healthy individuals use to limit the impact of upsetting events, reducing their
intrusiveness (Anderson and Hanslmayr 2014). It may also be the key pro-
cess underlying the previously discussed phenomenon of mnemic neglect
(Sedikides et al. 2016).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 151
Reconsolidation External
Extinction
Retrieval environment
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
152 R. M. Visser et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 153
Several observations, however, are incompatible with the view that sup-
pression is maladaptive. First, suppressing unwelcome thoughts is a wide-
spread behavior that serves many different functional purposes, as outlined
above. Simply concentrating on an important task, by its nature, requires that
an individual excludes distracting content. Second, achieving emotional bal-
ance after an upsetting event involves “getting over” unwelcome thoughts and
feelings, a process that requires active regulation of thoughts. People who have
difficulty with upsetting occurrences (e.g., getting over arguments or perceived
slights) are viewed as coping less well than people who quickly recover. Third,
after a trauma, most people initially experience intrusive thoughts and memo-
ries that usually diminish over time. People who cannot reduce their intru-
sions are classified as having mental health concerns; those who are capable
of reducing intrusive thoughts effectively are considered resilient, a clearly
positive attribute. Fourth, people with anxious thoughts about a feared event
that they cannot discard are considered as less healthy, compared to those who
are able to set aside fears and cope well. Finally, the proposal that trying to
stop unwanted thoughts is intrinsically ineffective is oddly discordant with the
massive literature on inhibitory control and, more broadly, attentional control,
in which controlling cognition and behavior is not only desirable but an essen-
tial capacity of intelligence. Indeed, the same clinicians that classify suppres-
sion as maladaptive often maintain beliefs that are contradictory to that stance
upon closer inspection. For example, the belief that attention bias modification
(i.e., training people to ignore unpleasant interpretations) is a desirable therapy
amounts to a belief that it is desirable to become skilled at ignoring unpleasant
contents, exactly what people do through suppression. The belief that people
with intrusive thoughts suffer from cognitive control deficits that make them
vulnerable to such thoughts assumes that resilient people can set aside upset-
ting thoughts, which must therefore be beneficial.
What do we make of these contradictions? What accounts for the view that
suppression is maladaptive, given its ubiquity and utility? One reason derives
from work on thought suppression using the white bear paradigm (Wegner
et al. 1987). In this paradigm, participants are instructed not to think about a
particular thought (e.g., a white bear) over a five-minute period. During that
time, if they happen to think about the thought, they were asked to report this.
Afterward, they are given a five-minute free expression period. Wegner found
that participants asked to suppress thoughts of a white bear ironically experi-
enced more white bear thoughts than participants who were not asked to sup-
press. This enhancement of the unwanted thought after suppression was termed
the rebound effect (Wenzlaff and Wegner 2000). Across nonclinical samples,
meta-analyses show that people exhibit a small to medium rebound effect com-
pared to control instructions (Abramowitz et al. 2001). Wegner and Zanakos
(1994) also introduced the White Bear Thought Suppression Inventory, a scale
intended to identify people prone to thought suppression. This instrument be-
came widely used in clinical research and was found to correlate with both
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
154 R. M. Visser et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 155
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
156 R. M. Visser et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 157
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
158 R. M. Visser et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 159
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
160 R. M. Visser et al.
to a particular topic (e.g., danger and threat in PTSD and anxiety; substance-
related information in substance use disorders).
Substantial research has focused on cognitive control processes, including
the role of inhibitory control mechanisms in intrusive thinking. Memory con-
trol mechanisms related to action control are known to suppress retrieval of
specific, episodic long-term memories and can work more globally to block all
episodic memory retrieval in response to a cue. Alternatively, reminder cues
can be used to selectively retrieve alternative distracting content, suppress-
ing the intruding thought via the mechanisms of retrieval-induced forgetting.
Moreover, retrieving information may trigger destabilization and/or restabi-
lization of a thought. As such, retrieval provides an opportunity not only to
increase the strength of a memory, but also to reduce it. In addition, working
memory processes are critically linked to the encoding and retrieval of long-
term memory and are important to guiding the selection of information for
consolidation into, and retrieval from, long-term memory (see Figure 9.1). It
seems likely that for an intrusive thought to be experienced, it would enter
working memory, making this system central to having an experienced (con-
scious) mental event.
The previous section described examples of research into the mechanisms un-
derlying intrusive thinking and their control using a variety of methods. Here
we present an overview of the methods and paradigms that are most commonly
used to study intrusive thinking in humans, nonhuman animals, or both, each
with its own strengths and weaknesses.
• Experience/thought sampling is a form of self-report that is of-
ten used to assess mind wandering in the present moment (Larson
and Csikszentmihalyi 1983; Bolger et al. 2003; McVay et al. 2009;
Schooler et al. 2011; Baird et al. 2014; Fraley and Hudson 2014). One
form has people pressing buttons when they catch themselves mind
wandering or experiencing other intrusions. This can be done in the lab
or in real life (e.g., using a diary or smartphone). Alternatively, one can
also probe people at irregular intervals to report whether they were on
task or not. A comparison of probed versus unprobed self-report shows
that having an experience and knowing that you are having an experi-
ence are different things (Schooler et al. 2011; Baird et al. 2014). The
first requires metacognition and this is imperfect, and possibly worse
in clinical populations.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 161
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
162 R. M. Visser et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 163
Affective Biases
A bias toward thinking about negative events can arise through utility-
weighted sampling, a process whereby the samples (in this case, what we
choose to think about) are not just proportional to the probability of the event
but also to its importance. Humans routinely overestimate the probability of
extreme negative events (Lichtenstein et al. 1978), and such biases have long
been viewed as a signature of irrationality. However, they may alternatively
reflect a rational use of limited resources. When many different outcomes are
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
164 R. M. Visser et al.
(a) (b)
(c) (d)
+140
2 1
Action 1
–7
–2
Action 2 0 0
+2
0
Action 3
p(value)
3 6
4 5
value
Figure 9.2 Computational approaches to thought guidance: (a) The meta-reasoning
problem of optimally apportioning limited resources can be formalized as a decision-
making problem over decision-making problems. While the standard decision-making
problem is the tree at the top without thick lines, the meta-reasoning problem is a tree
of such trees, where each branch corresponds to choosing to evaluate one branch (thick
lines) in the original decision problem. Intrusions could relate to a tendency to only
sample one set of options (arrows). (b) Prioritized replay: In this state-space, if the
reward (Euro coin) is received in the bottom right corner, then standard model-free
learning updates only the state immediately adjacent to it. Prioritized replay allows
memories to be reused multiple times to update more distant states; here, propagating
the information about the reward all the way back to the starting state. If the need or
gain functions determining what memories to replay are altered, then this could result
in repetitive replay of the same memory (arrows) as well as a failure to update dis-
tance states (i.e., to integrate the memory with other memories). Adapted after Mattar
and Daw (2018). (c) Guidance of thoughts by habits: Consider the situation where a
valuation system provides distributions for likely values of actions. Here, action 1 is
clearly inferior to the other two, and it appears that action 2 is the best. If the agent
were to invest computational effort into refining the estimates of the values of the ac-
tions, it would be best to examine action 3, as it may be even better than action 2. In
this manner, a goal-directed or model-based system could elaborate on approximations
provided by a simpler valuation system. This would also mean that the goal-directed
thought choice could be (mis)guided by habits. (d) Stopping aversive thoughts: In this
task, participants are extensively trained to learn to navigate a maze where each transi-
tion incurs some gains or losses. When given the opportunity to plan freely, they stop
internal simulations when they encounter one of the salient losses in red. Adapted after
Huys et al. (2015).
possible, it becomes difficult to consider them all, and humans are thought to
simulate instead a few outcomes in their mind and average the results of these
few simulations. If these simulations are too few in number, they are likely
to miss very important outcomes. Simulating in proportion not only to the
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 165
probability, but also in proportion to the utility, allows the resulting estimates
to be more robust. That is, the apparent overestimation of extreme events may
help mitigate cognitive limitations and explains a number of apparent irratio-
nalities (Lieder et al. 2018b). Applied to the setting of intrusive thoughts, it
provides one argument for why salient negative events should be simulated
even when they are very unlikely. In fact, the more aversive the event, the
higher the likelihood of simulation, suggesting one path by which the per-
ceived negative valence of a simulated event might increase the frequency
with which it is thought about.
Trauma Replay
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
166 R. M. Visser et al.
the notion that excessive negative appraisals of the sequelae of a trauma might
relate to the emergence of PTSD symptoms (Ehlers and Clark 2000).
Obsessions
Rumination
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 167
et al. 2012, 2015). Computational models of the choices in this task suggest
that participants internally simulate potential routes through the maze and
terminate simulation when they encounter a salient loss. Functional mag-
netic resonance imaging suggests that this inhibition recruits the subgenual
anterior cingulate cortex (Lally et al. 2017), a region known to be impor-
tant in depression (Drevets et al. 1997), its treatment (Mayberg 2009), and
rumination (Hamilton et al. 2015). Indeed, individuals who score high on
self-reported rumination show reduced pruning; that is, a reduced tendency
to terminate thoughts when encountering large losses during their internal
simulations (Q. Huys, pers. comm.), suggesting that rumination might di-
rectly relate to an inability to inhibit aversive thoughts, possibly via impair-
ments involving the subgenual anterior cingulate.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
168 R. M. Visser et al.
tapping into quite different processes, the latter most obviously addressing
the contents of monitoring operations in meta-consciousness. This reliance
on meta-awareness can be problematic and may be especially compromised
in clinical populations. More importantly, in the field of intrusive thinking,
there is a clear need for the development of better self-report instruments:
at present, there does not appear to be a measure that captures all aspects of
intrusive thinking in one instrument. In particular, there are four gaps in this
field (for discussion, see Banich, this volume).
1. Most questionnaires were designed with a particular clinical syndrome
in mind (e.g., PTSD, OCD, craving). As such, a general intrusive think-
ing scale does not exist.
2. For the most part, the questionnaires do not assess both the content
of the thought as well as the ability to control those thoughts, broken
down by content.
3. There are few, if any, questionnaires that specify the quality of intru-
sions across numerous dimensions, such as their form (verbal, images,
urges), vividness, salience, and frequency.
4. There are no questionnaires that systematically assess how intrusions
are triggered or the context in which people experience intrusions.
Looking forward, future research should focus on designing a new, theoreti-
cally motivated questionnaire, refining existing objective measures and capi-
talizing on the potential of computational approaches.
Cognitive Control
Action Control
A critical prefrontal region for stopping movements is the right inferior frontal
cortex (rIFC) as shown by lesion and transcranial magnetic stimulation studies
(Aron et al. 2014). It is thought that the rIFC, in concert with the presupple-
mentary motor area, implements stopping through a (fast) hyperdirect pathway
to the subthalamic nucleus (STN) of the basal ganglia, and this suppresses tha-
lamic drive back to cortex, leading to movement cancellation (reviewed by Bari
and Robbins 2013; Jahanshahi et al. 2015; Wessel and Aron 2017). Importantly,
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 169
in the standard case, subjects apparently stop using a global mechanism. For
example, stopping the voice suppresses the hand (Badry et al. 2009; Cai et al.
2012). This broad skeletomotor suppression begins around 120 ms after the
Stop signal and lasts for 100 ms or more. It is thought that the broad suppres-
sion relates to a wider putative impact of the STN on basal ganglia output (Aron
2011), just as the degree of global motor suppression in the Stop-Signal para-
digm relates to the level of oscillatory power in the STN (Wessel et al. 2016).
Yet subjects can also stop selectively when they are forced to do so (Aron
2011). For example, they are able to stop one response (or hand) while continu-
ing with another with minimal interference, and this does not result in broad
motor suppression (Majid et al. 2012). It is possible that this form of selective
response suppression relates not to a hyperdirect cortico-STN connection but
rather to a frontal-striatal-pallidal, so-called, indirect pathway. For example, peo-
ple with degeneration of striatum and pallidum, who are thought to have an indi-
rect pathway disorder, could not stop as selectively in the paradigm and did not
show typical physiological signatures of selective stopping (Majid et al. 2013).
As will be discussed below, this global versus selective picture in motor
stopping may have relevance for global versus selective control over memory.
Future work could develop behavioral paradigms to look at this selective con-
trol on the analogy of selective response suppression, taking into account the
observation that selective stopping is best done when the subject proactively
sets it up ahead of time (Cai et al. 2011; Majid et al. 2012); that is, it could
be possible to go into a situation preparing to suppress a particular memory
intrusion.
Working Memory
The three working memory operations that we discussed above may be par-
ticularly relevant to intrusions:
1. Gain access or gate information into working memory.
2. Select information within working memory to be given priority.
3. Remove information from working memory.
In general, the neural correlates of working memory mechanisms have been
well described (D’Esposito and Postle 2015) and likely involve prefrontal ar-
eas that are involved in executive control processes (e.g., selecting among in-
formation in working memory), basal ganglia mechanisms that work to gate
information into working memory, and posterior brain regions that help to pro-
vide sensory or abstract (e.g., semantic) representations of information to be
activated and/or placed in working memory. Most of the paradigms utilized in
cognitive neuroscience require the confluence of processes whereby informa-
tion enters, is selected, and updated in working memory (e.g., the N-back task).
Nonetheless, the neural bases of some of these three main processes have been
distinguished.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
170 R. M. Visser et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 171
Less research has examined how one can take information currently held
within working memory and dispose of it completely. This has been a difficult
question to address because it is hard to verify that an item has indeed been
removed. Other than with behavioral methods such as self-report, which may
not be reliable, few (if any) methods are able to confirm that a thought has been
removed or inhibited. Yet understanding such processes could have important
implications for how interventions for intrusive thoughts might be created.
While there is little research in this area, some work has demonstrated that
brain imaging techniques can be used to help verify the removal of informa-
tion and simultaneously to examine the control mechanisms by which such
removal occurs. In one of the few studies of this nature, Banich et al. (2015)
utilized activity in the sensory cortex as a rough proxy for whether informa-
tion was currently active in working memory. With this approach, they verified
the presence of such activity when individuals were maintaining information
about a picture just viewed or a short tune just heard, or when they replaced
the item with something else. They also observed a lack of such activity when
the item was removed from current thought, both when participants specifi-
cally suppressed that item as well as when they did so by clearing their mind
of all thought. With regard to control mechanisms, a hierarchy of function was
observed. Common across the replace, suppress, and clear conditions—all of
which require a shift in attention away from the original item to something
else, compared to the maintain condition—was activation in superior pari-
etal regions implicated in shifting attention among items in working memory
(Koenigs et al. 2009). When information had to be removed from working
memory (the suppress and clear conditions) as compared to when an item was
present in working memory (the maintain and replace conditions), activation
was observed over regions of lateral prefrontal cortex involved in executive
processes that act on working memory. Finally, for the clear condition as com-
pared to all the other conditions, there was increased activation in the insula,
suggesting a shift of attention to bodily states (Craig 2011) as well as activa-
tion in inferior parietal cortex, which may represent a mechanism for altering
the bottom-up salience of information (Cabeza et al. 2008), either by reducing
the salience of visual information or by increasing salience of information de-
rived from bodily states (see also Fedota and Stein, this volume). In relation
to long-term memory, the global “clear” condition on the contents of working
memory may be analogous to a global stopping process that inhibits retrieval
of all information from the hippocampus, whereas the “suppress” condition
may be analogous to suppression of a specific memory, as discussed in more
detail below.
In summary, the brain regions relevant to working memory processes in-
volved in intrusive thought likely involve (a) content-specific cortical regions
needed to access the representation of information underlying the intrusive
thought: visual areas for visual images, language- or semantically related re-
gions for thoughts, limbic regions for emotional information, and hippocampus
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
172 R. M. Visser et al.
for episodic memory, (b), basal ganglia mechanisms that influence what gains
access to working memory (see Badre, this volume), and (c) prefrontal mecha-
nisms involved in control processes that select and manipulate information
within working memory or act to engender its removal.
An unresolved issue concerns the extent to which the processes deployed
in suppressing contents actively held in working memory are distinct from
or overlapping with those involved in suppressing the long-term memory re-
trieval of unwanted thoughts, prompted by reminders (see section on retrieval
suppression), and whether these different types of mental control tap into
unique processes that may be differently affected in psychiatric disorders.
Long-Term Memory
Memory Encoding
For over a century, it has been recognized that memories are initially un-
stable, subject to interference, until they are stabilized through a process of
consolidation (Ribot 1882; Muller and Pilzecker 1900; Burnham 1903). Once
consolidated, long-term memories are largely protected from interference,
save windows of destabilization that may occur upon retrieval (Sara 2000;
Nader 2003). Reconsolidation involves a relatively brief (few hours) cascade
of molecular and cellular processes enhancing synaptic efficacy via structural
changes; a longer process (days and weeks) involves system-level connectivity
changes between the hippocampus and cortical areas (McGaugh 2000; Dudai
2012). Emotionally significant experiences trigger the release of adrenal stress
hormones, stimulating norepinephrine in the amygdala, which in turn modu-
lates plasticity in the hippocampus, cortex, and other brain regions (Cahill et
al. 1994; Southwick et al. 2002; Cahill and Alkire 2003; Strange and Dolan
2004; Hurlemann et al. 2005). Abundant evidence links dysfunction in these
circuits to psychological disorders in which intrusions are a symptom. Figure
9.3 presents a schematic overview of the neural circuits underlying these major
domains and their possible (dys)function in intrusive thinking.
To the degree that encoding of information is tied to a specific event, such
as that often associated with PTSD, aberrant functioning in the emotion cir-
cuits as well as the hippocampal and medial temporal structures that support
episodic memory are likely involved (Liberzon and Martis 2006; Pitman et al.
2012). Evidence points to hyperactivity in the amygdala and dorsal anterior
cingulate cortex (the putative human homologue of the prelimbic medial pre-
frontal cortex), whereas the ventromedial prefrontal cortex and hippocampus
evince hypofunction, accompanied by impaired extinction learning and recall
(Milad et al. 2009; Milad and Quirk 2012; Logue et al. 2018). Most recently,
specific computations of threat, including learning parameters such as value,
prediction error, and learning rate, have been characterized in PTSD. It was
found that PTSD severity was related to overweighing of prediction errors
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 173
dlPFC vmPFC
Cognitive
regulation
Retrieval
suppression
Inhibitory control
Episodic (e.g., extinction)
memory
Hippocampus Context
Hypothalamus/brain stem
Emotional reaction
(e.g., freezing, sweating)
Figure 9.3 Most of what we know about the neural mechanisms of emotional learning
and memory comes from animal studies that utilize threat conditioning and extinction
as a model. Based on these studies, sensory information from neutral stimuli in the en-
vironment that reliably coincide with emotionally significant outcomes converge in the
lateral nucleus of the amygdala where associative learning occurs, conferring emotional
value on the conditioned cues. Projections from the lateral to the central nucleus engage
descending projections to the hypothalamus and brain stem, which mediate the expres-
sion of the conditioned response (e.g., freezing). Projections from the lateral to the basal
nucleus onto the striatum form a path that promotes active coping and goal-directed
behavior. The prelimbic region of the medial prefrontal cortex (mPFC) connects with
the amygdala to sustain the expression of emotional responses, while the infralimbic
region acts to counteract amygdala output and diminish emotional responses. As such,
the infralimbic mPFC (ventral mPFC in humans) is a critical region in the acquisition
and recall of extinction learning. The hippocampus exerts contextual control over the
expression of threat learning. The dorsolateral prefrontal cortex (dlPFC) is the region
involved in top-down cognitive regulation of emotional responses by influencing amyg-
dala via the vmPFC (Hartley and Phelps 2010; Schiller et al. 2010; Milad and Quirk
2012). The dlPFC also exerts top-down modulation of hippocampal activity to induce
retrieval suppression (Anderson et al. 2016).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
174 R. M. Visser et al.
al. 2013; Battaglini et al. 2016; Clark et al. 2016), highlighting that consolida-
tion processes are important in the formation of intrusive memories. Together,
these studies invoke neural mechanisms for the learning and flexible modula-
tion of emotional memories. Aberrant functioning of the salience and memory
circuitries might induce emotional inflexibility, lack of proper contextual con-
trol, and impaired ability to diminish inappropriate emotional reactions, allow-
ing strong negative memories to persist and intrude.
Similar mechanisms may come into play if a memory is associated with
particular actions or procedures, such as may occur in substance use disorders.
Here, the encoding likely involves basal ganglia systems and alterations in
specific cortical processes (Balleine, this volume), such as visual regions that
may become tuned to particular stimuli (e.g., paraphernalia associated with
substance use). Across disorders, when information is associated with emo-
tional salience or significance (e.g., salience of negative emotional information
in depression, association of loud sounds with a traumatic event in PTSD),
such encoding likely involves the amygdala, as well as the striatum, insula,
and dorsal anterior cingulate cortex (Fedota and Stein, this volume). However,
whether intrusive memories are stored in a qualitatively different way than
other emotional memories, or are merely strong memories and therefore more
accessible, is still a topic of debate (Berntsen and Rubin 2008, 2013; Brewin
2014; Bisby and Burgess 2017).
Retrieval Suppression
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 175
stopping involves stopping retrieval and not physical actions, a plausible sup-
pression target would be the hippocampus, a brain structure involved not only
in encoding new memories but also in their retrieval, at least for recently ac-
quired events and thoughts (Anderson and Green 2001; Anderson et al. 2004;
Anderson and Hanslmayr 2014).
Abundant evidence indicates that retrieval suppression, unlike motor re-
sponse inhibition, relies on prefrontally mediated downregulation of activity in
the hippocampus and other medial temporal lobe regions to stop retrieval, pre-
sumably by preventing pattern completion (Anderson et al. 2004, 2016; Depue
et al. 2007; Banich et al. 2009; Levy and Anderson 2012; Gagnepain et al. 2014,
2017; Benoit et al. 2015; Schmitz et al. 2017). Suppression reduces hippocam-
pal activation not merely relative to active retrieval in the Think condition, but
also relative to passive baseline conditions, and this negative BOLD response
arises from negative coupling between the right lateral prefrontal cortex and
the hippocampus, established using effective connectivity analysis (Benoit and
Anderson 2012; Gagnepain et al. 2014, 2017; Benoit et al. 2015; Schmitz et
al. 2017). Indeed, within-subject comparisons of action and memory stopping
establish a clear double dissociation, with retrieval suppression downregulating
the hippocampus more than action stopping, but action stopping downregulat-
ing motor cortical areas (M1) more than retrieval suppression (Schmitz et al.
2017). Stopping of unwanted memories and thoughts thus involves a distinct
frontohippocampal inhibitory control pathway that suppresses hippocampal
activity. Hippocampal suppression appears to be a blunt instrument that acts
globally on the hippocampal state. For example, when a person tries to suppress
retrieval to depress awareness of a particular unwanted thought, the forgetting
arising from that suppression is not limited to the item people intend to suppress;
rather, any other recently encoded memories that occur either before or after the
act of suppressing something are also forgotten, even if they are entirely unre-
lated to the content being suppressed (Hulbert et al. 2016). Thus, suppression
of unwanted thoughts induces an amnesic shadow in the temporal surround of
the suppression attempt, creating both anterograde and retrograde amnesia ef-
fects in healthy people. This finding has been linked to the global suppression
of hippocampal processes that not only stop retrieval but also disrupt encoding
and stabilization processes necessary to retain recent experiences (Hulbert et al.
2016). This global, systemic disruption of hippocampal activity is analogous to
the global stopping identified in motor response inhibition.
What do we know about how the prefrontal cortex achieves this form of
inhibitory control over hippocampal activity? Primate anatomical studies tell
us that top-down suppression of hippocampal activity is unlikely to be direct,
not only because there are no direct connections between the lateral prefrontal
cortex and the hippocampus (Anderson et al. 2016), but also because long-
range projections from the hippocampus are largely excitatory. To achieve an
inhibitory effect in the hippocampus, if the negative BOLD response is truly
inhibitory, the prefrontal cortex must drive local populations of inhibitory
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
176 R. M. Visser et al.
interneurons within the hippocampus to disrupt its function. Because all inter-
neurons in the hippocampus are GABAergic, this observation suggests that
individuals with higher concentrations of hippocampal GABA may show a
superior ability to suppress hippocampal activity by prefrontal influence.
Recently, Schmitz et al. (2017) found evidence of this, using a multimodal
imaging study that combined fMRI with magnetic resonance spectroscopy.
These findings indicate that people with higher concentrations of hippocam-
pal GABA showed greater downregulation during retrieval suppression, more
successful forgetting of intruding thoughts, and greater negative coupling be-
tween the right prefrontal cortex and the hippocampus. As such, local concen-
trations of hippocampal GABA may provide a pivotal function that enables the
prefrontal cortex to implement long-range inhibitory influence necessary for
control over intrusive thoughts. This discovery sheds new light on evidence of
diminished hippocampal GABA in many disorders characterized by intrusive
thoughts (Schmitz et al. 2017), which may be a heretofore unrecognized risk
factor in the pathogenesis of disordered control over intrusive thoughts.
Despite the unique hippocampal targets involved in implementing re-
trieval suppression, it is equally clear that both retrieval and action stopping
processes engage overlapping regions in the dorsolateral (BA 9/46/10) and
ventrolateral prefrontal (BA 44/45) cortex, suggesting the existence of do-
main general supramodal inhibitory control regions that may dynamically
recouple with task-specific target regions (Depue et al. 2016; Schmitz et al.
2017; Guo et al. 2018). These domain general regions are strikingly right
lateralized, strongly consistent with the long-standing claim by Aron et al.
(2004, 2014) that inhibitory control is right lateralized, although the regions
clearly include both dorsolateral and ventrolateral regions, and not simply
ventrolateral prefrontal cortex. These supramodal regions have been identi-
fied both through within-subjects conjunction analyses of action and retrieval
stopping (Depue et al. 2016; Schmitz et al. 2017) as well as conjunctions
performed on quantitative meta-analyses of independent studies from many
laboratories (Guo et al. 2018). Interestingly, action stopping and retrieval
stopping also appear to engage highly colocalized regions within the basal
ganglia (Anderson et al. 2016); for a detailed discussion of anatomical hy-
potheses, see Depue (2012), Guo et al. (2018), Paz-Alonso et al. (2013), and
Balleine (this volume).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 177
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
178 R. M. Visser et al.
working memory, however briefly. Some evidence supports this. For exam-
ple, Hellerstedt et al. (2016), using event-related potentials, found evidence
that the frontal negative slow wave (NSW), observed over the right prefron-
tal cortex, is modulated during intrusions. This component has been linked
in prior work to the storage of information in working memory. Consistent
with this, participants during Think trials show a prolonged NSW that lasts
throughout the full several seconds of the trial; in contrast, non-intrusions
show little evidence of this, consistent with the exclusion of items from work-
ing memory. Intrusions, however, showed a brief increase in the NSW, which
was rapidly eliminated within the first seconds of the trial, suggesting a brief
penetration of the intruding item into working memory. Perhaps relatedly,
Castiglione et al. (2019), using time frequency analysis, found evidence that
during No-Think trials, there is a robust increase in frontal beta component
during non-intrusions. Given the prior linkage of this component to motor
response inhibition, these findings suggest that intrusions may reflect an ini-
tial failure of inhibitory control that allows the intruding content to penetrate
working memory.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 179
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
180 R. M. Visser et al.
a bill) or solving a problem (an “aha” moment) (see also Monfils and Buss,
this volume). If mind wandering is a form of intrusive thinking, research
shows that the capacity to “jump out of the present set” can be adaptive and
relates to creativity. Its adaptiveness is likely reduced when mind wandering
becomes excessive or too diffuse, such as in ADHD, or when the content is
unpleasant.
The potential adaptive nature of intrusive thinking may be independent
of the desirability to control it. In healthy individuals, frequent and excit-
ing thoughts (e.g., love infatuations) might disrupt concentration on the
task at hand and therefore not be adaptive, yet still wanted in the moment.
Alternatively, intrusive thinking about mistakes that we have made may allow
us to adjust our behavior and become a better person, yet unwanted in the
moment. The discussion below on sociocultural implications provides com-
pelling examples of this. In addition, as a clinical symptom, intrusive think-
ing is not always unwanted (e.g., exciting flashforward thoughts that occur
during a manic episode), yet often maladaptive. Whether the opposite is also
possible (i.e., unwanted intrusive thinking in mental health disorders may in
some cases serve an adaptive purpose, such as in acute stress disorder as part
of processing trauma), is an open question that requires more research.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 181
Sociocultural Context
Intrusive thinking and its control (or lack thereof) take place within a sociocul-
tural context. Examples of this include the AIDS-HIV epidemic of the 1980s.
In his book, The Man Who Couldn’t Stop: OCD and the True Story of a Life
Lost in Thought, David Adam (2014) provides a vivid example of how frequent
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
182 R. M. Visser et al.
concerns about infection drove the induction of his own obsessive and com-
pulsive symptoms. Certainly, there are many other examples of widespread
general concern (e.g., nuclear war, Brexit, global pandemics) that intrude into
our everyday consciousness and may lead to pathological consequences.
Although many of us successfully resist such concerns, this too can lead
to societal consequences, as can be currently observed through the striking
example of the climate crisis. Since the Industrial Revolution, global tempera-
tures have increased by 1°C and humankind is well on track to experiencing an
additional increase of 0.5°C by 2030 (Xu et al. 2018). In line with the worst-
case scenario put forth by the Intergovernmental Panel on Climate Change
(IPCC 2018), this half a degree will likely correspond to a 50% increase in
extreme weather events worldwide (e.g., droughts, floods, snowstorms, hur-
ricanes, cyclones) and exact devastating consequences, including mass migra-
tion, agricultural failure, and deadly heat events (Xu et al. 2018). The near total
consensus of climate scientists, exemplified by the Paris Agreement, which
almost all governments signed (and almost none are honoring), is that we have
to reduce emissions soon. Failure to do so will result in continued temperature
increases that will soon be beyond human control and ultimately lead to a “hot-
house planet” by 2100, or perhaps even sooner (Wallace-Wells 2019).
The climate crisis is surely creating daily thought intrusions in hundreds
of millions of people. Such intrusions are likely characterized by interrupting,
salient, experienced events (imagery, emotion, moral feeling) that recur. This
example illustrates the complexity of judging whether intrusive thinking is mal-
adaptive or not, and the range of responses that people can have. For instance,
against the unimpeachable backdrop of scientific knowledge, such intrusions
appear highly adaptive: they compel action and yet only a very small minority
of people are currently engaged in action—the great majority of global citizens
are not taking action. Of these, some may deny the science or the predicted
impacts, or they may accept the science and impacts but deny that any serious
action is warranted (e.g., because they are ideologically committed to the cur-
rent economic system). In Western societies, polling shows that most people
fall in the latter category (accepting the science and the impacts), yet they are
not acting beyond some minor adjustments in their personal lives, possibly be-
cause action would be inconvenient to one’s career or lifestyle or because it
would require confronting grief and fear in a way that one is not yet prepared
to do. Since such people do understand and accept the terrifying imperative to
act, but are not doing so, they could be characterized as exerting control over
the intrusive thinking caused by the climate crisis. The form of control being
used is probably based on reappraisal: people express degrees of fatalism (“the
problem is too big or too hopeless,” “it is too late”), nihilism (“humans deserve
what is coming to them”), a deferral of responsibility to policy makers (“it is
a government problem”), or presently unwarranted faith in technology alone
to solve the problem (Hansen 2018). Nonetheless, as more extreme weather
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Neuropsychological Mechanisms of Intrusive Thinking 183
events, for instance, increase over time, so too will levels of anxiety, until the
intrusion breaks through the threshold of control and people clamor for action.
In summary and in accordance with the definition of intrusive thinking as
interrupting, salient, experienced events (imagery, emotion, moral feeling) that
are also recurrent, the climate crisis is clearly generating intrusive thinking in
a wide range of people around the world. This, in turn, creates conflicts which
can inculcate attempts at control (e.g., at the reappraisal level). Such tensions
exacerbate poor mental health as well as the feelings of instability and fear,
which are already driving populist political regimes (Latour 2018). Meanwhile
fake news disseminated within the broader system functions to degrade the
salience signal that is currently essential to generate the type of population-
level intrusive thinking that would compel quick action. This sociocultural
example is interesting because it represents the flip side of many of the psy-
chiatric-related examples discussed in this chapter. Whereas intrusions were
often characterized as pathological, the climate crisis example demonstrates
how intrusive thinking is a good thing for our biosphere and civilization, and
attempts to control these intrusions are maladaptive. The survival of human
civilization depends on more intrusive thinking right now.
Concluding Summary
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
184 R. M. Visser et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
10
Networks Relevant to
Psychopathology and
Intrusive Thought
John R. Fedota and Elliot A. Stein
Abstract
Intrusive thoughts are regular occurrences in healthy cognition. Across a variety of
psychiatric conditions, however, such thoughts can become unconstructive and per-
severative. Failures in computations to estimate the salience of the content of these
thoughts are at least partly responsible for these clinically relevant disease symptoms.
This chapter reviews neuroimaging results that show specific and related dysfunction
in the calculation of salience at multiple neuroanatomically and functionally linked
regions of interest, both cortically and subcortically. Transdiagnostic evidence for dys-
function in the striatum, thalamus, and prefrontal cortex is reviewed, as is a theoretical
framework placing these regional findings in the context of large-scale brain networks.
It is argued that changes in nodal function and network communication are signatures
of a failure to properly shape predictions about the reliability and utility of external
and internal stimuli, leading to maladaptive attentional capture and behavior, including
intrusive thoughts.
Introduction
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
188 J. R. Fedota and E. A. Stein
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Networks Relevant to Psychopathology and Intrusive Thought 189
governed by the relative precision of each (Mathys et al. 2011). That is, pre-
cisely estimated priors are less susceptible to modulation based on sensory
evidence, while poorly estimated (imprecise) priors are readily adjusted by
sensory evidence. As different sensory inputs can vary in their precision and
information content, one cognitive imperative is to estimate which available
sensory input will provide reliable and informative information to calculate
better posterior estimates of the environment (Parr and Friston 2017, 2019).
To this end, unambiguous sensory data should be amplified when present and
sought when absent.
Cognitively, this predictive estimation of potentially reliable sensory
sources has been described as the attribution of salience (Parr and Friston
2019). By salience, we mean a quality that is particularly noticeable or deemed
important to the individual in a given context (Uddin 2014; Kahnt and Tobler
2017; Miyata 2019). As such, elements that provide unambiguous sensations
should be ascribed higher salience, as they will provide reliable information
for the adjustment of posterior estimates and iteratively improve subsequent
decisions. The degree to which new information alters posterior estimates as
compared to prior beliefs is termed Bayesian surprise (Itti and Baldi 2009;
Barto et al. 2013).
Salience is closely related to value, but a key distinction is that value is a
signed currency that varies monotonically from negative to positive whereas
salience is an unsigned currency, where both negative and positive predicted
outcomes can have equivalent salience (Kahnt and Tobler 2017). This is be-
cause both positively and negatively valent elements of the environment can
improve posterior estimates; each can be informative in the refinement of pos-
terior estimates.
When working properly, this iterative cycle of hypothesis (prior)–result
(sensory evidence)–conclusion (posterior) allows for the flexibility and learn-
ing characteristics of human cognition. However, in the case of NRTs, im-
proper predictions of the salience of elements in the environment will lead
to suboptimal processing, including perseveration on elements with overly
precise priors and/or a failure to guide attention to elements with weak priors.
For example, if previous experience creates an overly precise prior belief
about the reliability of information to be gleaned from a stimulus (e.g., a drug cue
or emotionally valent memory), its salience will be increased. Such an increase in
anticipated salience will lead to the focus of greater attentional resources on the
stimulus, potentially at the expense of alternatives in the environment. This dys-
regulated focus on one thought at the expense of others is a hallmark of intrusive
thought, as described in the following sections. Indeed, the computational frame-
work of Bayesian predictive coding has been shown to be useful in describing
specific deficits across a variety of psychiatric conditions associated with NRTs:
hallucinations in SZ (Sterzer et al. 2018), drug cravings in SUD (Gu and Filbey
2017), and perseverative focus in OCD (Levy 2018).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
190 J. R. Fedota and E. A. Stein
A large body of literature has shown the striatum to be responsive not only to
value and reward, but also to the salience of a given stimulus, independent of
its positive or negative valence. In humans, these reward (magnitude * signed
valence) and salience (magnitude * absolute value of valence) responses are
separable: salience-evoked activation is seen in the ventral striatum (Zink et al.
2006; Jensen et al. 2007; Bartra et al. 2013), the insula, and the dorsal anterior
cingulate cortex (dACC), while reward (positive valence) is also encoded in
the striatum and across various brain regions, including the orbitofrontal cortex
(OFC) (Litt et al. 2010; Bartra et al. 2013; Kahnt et al. 2014).
Dysfunction in striatal signaling related to the identification of salient
environmental stimuli is observed across psychiatric conditions associated
with NRTs. When compared to healthy individuals, those at risk for devel-
oping psychosis show heightened activation to task-irrelevant stimuli within
the ventral striatum (Roiser et al. 2012; Schmidt et al. 2016). The observed
increase in ventral striatal activation to irrelevant stimuli suggests an over-
sensitivity to uninformative cues in the environment. Within the Bayesian
predictive coding framework described above, during normal cognition these
irrelevant cues should be ascribed reduced salience, as they provide little to
no sensory information with which to refine posterior estimates of the task
environment.
Similar biases in ventral striatal activation are seen in other conditions such
as OCD and bias in processing losses (Jung et al. 2011), MDD and anhedo-
nia (Whitton et al. 2015), as well as SUD and cue reactivity (Volkow et al.
2010; Kühn and Gallinat 2011). Further, in a recent review of SUD, Moeller
and Paulus (2018) observed that ventral striatal activation patterns are related
to long-term abstinence outcomes: increased activation in response to drug
cues is related to worse clinical outcomes (i.e., increased substance use and
recidivism), whereas increased activation to monetary or nondrug reward cues
are related to better clinical outcomes. In each case, ventral striatal activa-
tion is biased in its sensitivity to a variety of environmental cues. Thus, the
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Networks Relevant to Psychopathology and Intrusive Thought 191
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
192 J. R. Fedota and E. A. Stein
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Networks Relevant to Psychopathology and Intrusive Thought 193
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
194 J. R. Fedota and E. A. Stein
With NAc core dysfunction now described along with neuroanatomical links
between this region and thalamic and prefrontal regions, including the dACC
and aIns, our focus shifts from striatal to cortical areas implicated in psy-
chiatric disease associated with NRTs. A recent meta-analysis of structural
(n > 15,000) (Goodkind et al. 2015) and functional (n > 5,000) (McTeague et
al. 2017) data from psychiatric patients across a variety of disorders associated
with intrusive thoughts (SZ, MDD, SUD, OCD) identified specific yet over-
lapping areas of dysfunction. Specifically, the pattern of gray matter loss in
patients was circumscribed to dACC and bilateral aIns (Goodkind et al. 2015).
In agreement with these structural findings, hypoactivation in cognitive control
task-evoked activity was also observed in the dACC and right aIns as well as
in the left dlPFC and right IPL (McTeague et al. 2017). The tasks employed
in the meta-analysis did not probe intrusive thoughts per se, but rather top-
down cognitive control more generally. That said, the pattern of hypoactivation
across disease conditions showed reductions in regions associated with both
attentional control (i.e, left dlPFC, right IPL) and the calculation of salience
(i.e., dACC, aIns).
The aIns and dACC are coactivated across a wide variety of cognitive con-
trol and attention-related tasks. In fact, activation in these nodes are among the
most observed results in the fMRI cognition literature (Behrens et al. 2013).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Networks Relevant to Psychopathology and Intrusive Thought 195
Given their ubiquity in the extant literature, aIns and dACC have been theo-
rized to play a central role in broad cognitive constructs, as in the detection
of relevant information from both the external and interoceptive worlds and
the coordination of appropriate attentional capture and behavioral response to
these salient signals (Uddin 2014; Heilbronner and Hayden 2016; Nour et al.
2018). The degree of task activation in these regions increases with the demand
for attentional control or with an increase in ambiguity of stimuli during per-
ceptual decision making (Lamichhane et al. 2016).
The insula is associated with the integration of interoceptive information
into calculations of salience (Critchley et al. 2004; Craig 2009) via ascending
pathways communicating visceral and allostatic information (Critchley and
Harrison 2013; Kleckner et al. 2017). Within the insula (and the dACC, as dis-
cussed below), these signals are integrated to create a single subjective image
of “our world” (Kurth et al. 2010); interoceptive representations in the insula
have been theorized to provide the basis for a perception of self via the inte-
gration of interoceptive signals related to physical state (Seth 2013; Namkung
et al. 2017). This self-focused processing localized to the insula is strikingly
consistent with the description of the content of repetitive thought as being
“related to one’s self and one’s world” (Segerstrom et al. 2003).
Returning to a transdiagnostic theme and the observation of insular hypo-
activity (McTeague et al. 2017), impairments in interoception are implicated
across psychiatric conditions associated with NRTs (Khalsa et al. 2018). Insular
activation in response to cued recall of a previously interoceptive challenge was
diminished in subjects with MDD as compared to healthy controls (DeVille et
al. 2018). Similar impairment in interoceptive processing focused on insular
hypoactivation (Naqvi and Bechara 2010) has also been associated with SUD
(Goldstein et al. 2009; Sutherland et al. 2013; Paulus and Stewart 2014).
In addition to a central role in the integration of interoceptive signals, the
insula is a primary adjudicator between external (exogenous) and internal (en-
dogenous) focus as a function of salience attribution (Sridharan et al. 2008;
Menon and Uddin 2010; Uddin 2014). Characterization of the causal inter-
actions between insula and dACC, however, shows additional divergence in
their patterns of activation. Specifically, aIns has been shown to amplify the
detection of salience within the dACC (Chen et al. 2014; Cai et al. 2015). This
timescale is consistent with EEG spectral analysis showing insular activation
precedes that of dACC (Chand and Dhamala 2016).
In contrast to the association with interoceptive processing in the insula,
the dACC is associated with cognitive monitoring and control processes along
with economic decision making (Botvinick 2007; Kolling et al. 2016; Shenhav
et al. 2016; Alexander and Brown 2017). Recent integrative accounts suggest
that the primary function of the dACC is to process multiple facets of informa-
tion about the context in which a decision is being made to enable the appro-
priate goal-directed strategy (Heilbronner and Hayden 2016; Li et al. 2018b).
That is, dACC codes task-state information (originating either endogenously
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
196 J. R. Fedota and E. A. Stein
Within the described CSTC loop, dysfunction and/or maladaptive bias to-
ward potentiated stimuli is observed at multiple levels of salience processing.
Additionally, many of the implicated regions of interest are regularly activated
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Networks Relevant to Psychopathology and Intrusive Thought 197
in concert depending on task demands. For example, Nour et al. (2018) show
ventral striatal, dACC, and aIns activation in the processing of informative,
novel information, with the dACC and aIns regularly coactivated (Behrens et
al. 2013). These commonalities suggest a benefit to examining the brain at a
higher level of organization, moving from regions of interest to dyadic circuits
to large-scale network organization.
To this end, dACC and aIns are the primary nodes of the salience network
(SN). Originally described by Seeley et al. (2007), and subsequently replicated
using a variety of methodologies (Dosenbach et al. 2007; Smith et al. 2009;
Power et al. 2011), the SN is a centralized processor that ascribes salience
to stimuli (Uddin 2014) while coordinating attentional resources between an
internal and external focus in response to task demands (Sridharan et al. 2008;
Menon and Uddin 2010). Such a function is clearly consistent with the roles
ascribed individually to aIns and dACC detailed above.
While the dACC and aIns form the primary nodes in the SN, additional
large-scale networks are broadly relevant to cognition (Smith et al. 2009; Ji et
al. 2019). The interaction between the SN and these networks leads to a more
global view of the dysfunction associated with intrusive thought; such SN in-
teractions can be conceptualized within a tripartite network model. Briefly, this
model describes a default mode network (DMN), which includes the rostral
and posterior cingulate cortices, parahippocampal gyrus, and bilateral infe-
rior parietal cortex (Raichle et al. 2001; Buckner and DiNicola 2019), and an
anticorrelated executive control network (ECN), which includes the bilateral
dlPFC and parietal cortices (Honey et al. 2007). During interoceptive process-
ing, the DMN is relatively more active and the ECN is relatively deactivated
(Fox et al. 2005; Keller et al. 2013). During exteroceptive processing the re-
verse is true. Toggling between these two networks is thought to be mediated
by the SN (Menon and Uddin 2010).
Indeed, regardless of the cognitive function ascribed to the individual re-
gions of the SN, transdiagnostic findings implicating nodes of the SN, both
structurally (Goodkind et al. 2015) and functionally (McTeague et al. 2017),
have led to conceptualizations of SN dysfunction as a core transdiagnostic
symptom of psychiatric dysfunction (Menon 2011). Transdiagnostic differ-
ences in activation within nodes of the CSTC loops connecting striatum to SN
nodes have been described in detail by Peters et al. (2016).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
198 J. R. Fedota and E. A. Stein
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Networks Relevant to Psychopathology and Intrusive Thought 199
boundaries, when the identified occipital regions are used as seeds in a rest-
ing-state functional connectivity analysis, similar to those analyzed by Sha et
al. (2019). Nonetheless, hyperconnectivity with both the ECN and DMN was
positively correlated with p factor score. Importantly, while DMN and ECN
within the tripartite network model are usually considered oppositional (Fox
et al. 2005; Sridharan et al. 2008; Menon and Uddin 2010), the connectivity
of both networks was similarly enhanced with an independent, and psychiatri-
cally relevant, region of the occipital cortex. Further, the degree of this shared
increase in functional connectivity correlated positively with p factor score.
These results point to network interactions beyond the SN (Peters et al.
2016) that may indirectly influence the salience calculations instantiated within
dACC and aIns. In both of these recent cases, the network structure separat-
ing ECN and DMN appears to be reduced, either through hyperconnectivity
between these two normally oppositional networks (Sha et al. 2019), or via
increased coherence with a mediating node outside of the tripartite network
(Elliott et al. 2018). The coherence between ECN and DMN may further bias
the information integrated in SN, though an empirical demonstration of this
remains outstanding.
Conclusion
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
200 J. R. Fedota and E. A. Stein
well-described CSTC loops (Choi et al. 2017) to the thalamus, where recent
evidence suggests their representations in PFC may be amplified or sustained
(Parnaudeau et al. 2018; Pergola et al. 2018), potentially increasing bias.
Hypoactivation of nodes within the SN (Peters et al. 2016; McTeague et
al. 2017) leads to suboptimal integration of interoceptive signals (Kleckner et
al. 2017), which may be biased due to the allostatic load of psychiatric dis-
ease more generally (McEwen and Gianaros 2011). In addition, the integra-
tion of endogenous and exogenous information to identify and fully process
contextually relevant information (Heilbronner and Hayden 2016) is likely
biased by the observed hypoactivity within nodes of the SN. These process-
ing biases within SN lead to an inability to identify relevant information in
a given context (Parr and Friston 2019) or to update beliefs to modify atten-
tional strategies or behavior more generally (Nour et al. 2018). Finally, the
integrative calculations of the SN may be further impacted by alterations in
large-scale network structure (Sha et al. 2019), which further bias salience
calculations by reducing the fidelity of information communicated with the
rest of the brain.
In summary, bias at each stage of salience attribution leads to an over-
representation of potentiated stimuli as well as to an insensitivity to counter-
factual evidence, which normally signals the need to alter behavior. A better
understanding of the calculations instantiated within each of these regions,
and more importantly, a more holistic, systems-based picture of their interac-
tions, is likely to identify novel therapeutic interventions that will allow us to
mitigate the unconstructive consequences of NRTs and to treat the underlying
dysfunction.
Open Questions
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Networks Relevant to Psychopathology and Intrusive Thought 201
Acknowledgments
This work was supported by the National Institute on Drug Abuse, Intramural Research
Program and Center for Tobacco Products (U.S. Food and Drug Administration) Grant
No. NDA13001-001-00000 (to EAS).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
11
David Badre
Abstract
The last decade has witnessed a marked shift of emphasis in cognitive neuroscience
away from simple localization of function and toward the organization, coding, and
dynamics of brain networks. This is surely a healthy evolution of our science, and the
study of cognitive control has benefited from this shift, as much as any domain. Howev-
er, the emphasis on brain-wide networks for cognitive control has reopened some older
debates, once thought resolved, while also introducing some new ones. This chapter
focuses on four questions viewed as unresolved and fundamental because one’s par-
ticular answer to them commits to some basic theoretical differences regarding cogni-
tive control function: Are there one, many, or any networks whose primary function is
best described as cognitive control? Are the networks supporting cognitive control in
the brain “hub-like” or “hierarchical” in their intrinsic and extrinsic organization? Are
the networks for cognitive control modulatory or transmissive in the pathway from
thought to action? Does controllability apply at the level of cognitive function or brain
state? Each question is defined and relevant background is presented that could inform
a resolution.
Introduction
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
204 D. Badre
the fly, and to adapt to open-ended problems and novel situations. It allows
us to sustain goal-directed behavior over multiple timescales and to withhold
inappropriate responses, even when those responses are prepotent, habitual, or
stem from the prevailing urges of the moment. Cognitive control function lies
close to the heart of human intelligence and ingenuity. It is also vulnerable to
deficits across many, if not most, psychiatric and neurological disorders, be-
ing at the base of many of the behavioral problems arising in those conditions.
Thus, understanding the mechanisms by which the brain supports cognitive
control is a problem of fundamental importance.
Understanding cognitive control is of direct importance for intrusive think-
ing, the definition and scope of which is addressed in detail in other chapters
of this volume. Most definitions, however, require that intrusions are unwanted
and are unrelated to our goals or the task at hand. Thus, control mechanisms
are an important means by which we both avoid intrusive thoughts and man-
age their impact. It follows that understanding the brain systems that support
cognitive control function will have important implications for intrusive think-
ing, both in identifying its sources and seeking its potential remediation. In this
chapter I review the brain networks that support cognitive control as a general
background for more direct consideration of intrusive thinking.
As with most domains of cognitive neuroscience, the last decade of research
into cognitive control in the brain has witnessed a shift away from a paradigm
of functional localization toward one of functional networks. Among the most
robust and important observations to emerge from the overall network approach
has been that sets of brain areas tend to covary mostly with each other and not
with other areas (Power et al. 2011; Yeo et al. 2011; Buckner et al. 2013). Further,
the structure of this covariation is not entirely due to spatial proximity. Rather, af-
filiated areas can be distributed in each lobe of the brain, whereas other areas that
are spatially contiguous may not affiliate. These basic properties have allowed
for definition of brain networks or clusters of areas that covary with each other at
different scales (Power et al. 2011; Yeo et al. 2011).
Here, I focus on four big questions that are provoked when one takes a net-
work view of cognitive control seriously:
• Are there one, many, or any networks whose primary function is best
described as cognitive control?
• Are the networks supporting cognitive control in the brain “hub-like”
or “hierarchical” in their intrinsic and extrinsic organization?
• Are the networks for cognitive control modulatory or transmissive in
the pathway from thought to action?
• Does controllability apply at the level of cognitive function or brain state?
Obviously, this is not intended as an exhaustive list of questions about control
networks. Rather, these are the kinds of questions that I find myself asking
routinely, whether in my own work or in reading about others’. No one has de-
finitive answers, and so these questions also remain contentious or unresolved.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Brain Networks for Cognitive Control 205
My goal is not to provide answers in this essay, though I will express my own
view. Rather, I will define each question and present some relevant background
in the hope of provoking further discussion.
One of the oldest questions in the study of cognitive control or executive func-
tion is whether there exists one or many executive controllers in the mind and
brain, or if there are executive controllers at all. The majority view has mostly
been that, while there exists cognitive control function, it is not simply one
thing. Rather, what we call executive function or cognitive control actually
refers to a variety of specific functions and capacities.
Two camps reject this basic view. First, there are those who contend that
there is one central system for cognitive control or executive function and that
little to no decisive evidence exists for strong dissociations among subtypes of
cognitive control functioning. The second camp argues that control is emergent
from network processing in the brain, but that no particular area or network
of areas is best characterized as primarily supporting “control.” Finally, even
among those who agree that cognitive control exists and has many facets, there
has been little agreement about the exact type and number of these facets.
This core debate has unfolded in almost every domain in which cognitive
control has been studied: from behavior to individual differences to neuropsy-
chology to neuroimaging. Currently, it is playing out again in network neu-
roscience. I will devote some more space to this first question than the other
questions as it also provides an opportunity to summarize some background on
the networks relevant to cognitive control.
One reason that the unitary hypothesis has been so hard to falsify conclusively
is that it is often the null hypothesis (Aron et al. 2015). It predicts that in any
setting in which one attempts to locate a difference based on a type of cogni-
tive control, there will be no difference. Thus, any imprecision in design, logic,
or measurement has the potential to find evidence consistent with this unitary
view by virtue of being inconsistent with the alternative. As a consequence,
the unitary view has been something of a “zombie hypothesis” over the years:
falsified in experiments that show dissociations in the brain or behavior, only
to rise again a few years later when the same distinction is not found to gen-
eralize to a new task or the methodology changes. However, it is important to
acknowledge that a failure to locate a difference, even in direct replication, is
not itself positive evidence for a unitary controller. Rather, unitary controllers
need positive predictions and evidence of their own.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
206 D. Badre
In this light, the definition of the multiple demand system put forth by
John Duncan and colleagues is appealing as a unitary controller view of brain
organization, because it is based on a positive prediction: the multiple de-
mand system is engaged when you perform any challenging or difficult task
(Duncan 2013). Under these difficult circumstances, one needs to sequence
the set of attentional states required to perform the task. It is also in these
“hard” settings where one should expect the unitary cognitive control system
to be engaged. Importantly, however, the specifics of the task in question or
the demands that made the task difficult are not important. This system should
be fundamental and domain general, so that it participates across these differ-
ent task settings.
To test this hypothesis, Fedorenko et al. (2013) conducted an fMRI experi-
ment in which they contrasted difficult versus easy conditions in a wide range
of tasks. Difficulty was simply defined as a condition that took longer and
induced more errors behaviorally. The tasks differed in their specific demands
and the domain of input, such as between verbal or spatial. Nonetheless, when
one contrasted the hard with the easy conditions of these tasks, a consistent
set of areas was activated in each participant, as shown in Figure 11.1a. Given
its definition, this network was dubbed the “multiple demand system” or MD
system (Fedorenko et al. 2013).
The MD system has been studied extensively. It includes premotor cortex,
lateral prefrontal cortex (PFC) around the inferior frontal sulcus, the intrapa-
rietal sulcus, the anterior cingulate cortex (ACC), the frontal operculum, and
subregions of the basal ganglia (Fedorenko et al. 2013). This network has been
associated with a variety of measures of flexible behavior, including general
intelligence (Woolgar et al. 2010) and novel rule following (Tschentscher et al.
2017). In addition, most recently, it has been found to line up with the Human
Connectome Project parcellation that is defined based on a range of structural
and functional anatomical features (Assem et al. 2020).
As a unitary system, the MD system is proposed to serve a very gen-
eral control function needed across multiple complex tasks; namely, the
assembly of attentional episodes that are the smallest unit chunks of a
complex problem (Duncan 2013). When people seek to solve a new or
difficult task, it has long been thought that they must break the problem
into parts (Newell 1990). From the MD theory, each part is defined by a
set of input-output relations that are coordinated by attentional systems.
The MD system is proposed to manage these attentional episodes and the
transitions from one to the next. Thus, neural coding within this network is
thought to be highly dynamic, changing from moment to moment in a tra-
jectory determined by the flow of attentional episodes. The consistent and
widespread observation of flexible and dynamic coding of prefrontal neu-
rons from electrophysiological recording in the nonhuman primate shares a
qualitative correspondence to this view of multiple demand coding (Rainer
et al. 1998; Stokes et al. 2013).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Brain Networks for Cognitive Control 207
(a)
Premotor cortex
Intraparietal Pre-SMA / anterior cingulate
sulcus
Inferior frontal
sulcus
Anterior insula /
frontal operculum
(b)
CO
FP
Figure 11.1 Networks activated across multiple task demands. (a) Activated regions
of multiple demand systems: contrast of hard versus easy conditions in all tasks run
(after Fedorenko et al. 2013). (b) Frontoparietal (FP) and cingulo-opercular (CO) net-
works defined through functional connectivity: different methods of network definition
find convergent network definitions (after Gratton et al. 2018).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
208 D. Badre
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Brain Networks for Cognitive Control 209
Within the broadly defined FP network there is evidence for further func-
tional distinctions and subnetworks (e.g., Dixon et al. 2018). These distinc-
tions have been most consistently observed in the context of complex tasks
that are designed to test hierarchical cognitive control (Badre and Nee 2018).
Hierarchical cognitive control refers specifically to cases where we must con-
trol actions based on immediate contextual signals, while also being influenced
by higher-order superordinate control signals that are either more abstract pol-
icy or extended in time.
In general, if a task requires tracking multiple contextual signals to keep over-
lapping behavioral policies separate, demands on hierarchical control grow. For
example, in a recent experiment, children and adults were instructed with a set of
mappings between cartoon characters and left or right button presses, the “Go”
task (Verbruggen et al. 2018). Prior to performing the Go task, however, partici-
pants were asked to view all of the cartoon characters, pressing the right button to
advance to the next character (the “Next” task). This meant that while perform-
ing the Next task, participants would occasionally press the rightward arrow to a
character that required a left response later on during the Go task. Such an over-
lap of responses can cause conflict, evidenced in slowed response time during the
Next task. However, this conflict, is reduced if one can impose a latent context
that separates the episode of the Next task from the later episode of the Go task
and their respective response sets.
Interestingly, when doing the Next task, children exhibited more conflict
than adults; children had a harder time imposing this context episode on the
task. Notably, this conflict was evident even though they had never performed
the Go task and were only instructed on the response rules for this task. So,
it was not rule following that was a problem for the children, perhaps counter
to the widely held view. The conflict indicates they immediately implemented
the rules just from the instruction. Rather, their slow response was a symptom
of diminished hierarchical control capacity: they could not keep the latent task
contexts separate.
Studies testing hierarchical control have consistently exposed differences
within the FP control system (Figure 11.2a). Across a range of studies us-
ing fMRI, transcranial magnetic stimulation (TMS), and testing of patients
with frontal lobe lesions, differences in policy abstraction (defined in terms
of the number of conditions or branches in a decision tree between stimulus
and response) yield differences along the caudal to rostral PFC, with the
highest levels of abstraction associating with the rostral mid-dorsolateral
PFC (Koechlin et al. 2003; Badre and D’Esposito 2007, 2009; Nee and
D’Esposito 2016, 2017; Badre and Nee 2018). Further, manipulations of
temporal abstraction, which refers to the degree to which a goal or task must
be held pending over time, have found fMRI activation in the most rostral
portion of the frontal cortex, the rostrolateral PFC (Koechlin and Hyafil
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
210 D. Badre
(c)
0.5 2
2007; Desrochers et al. 2015; Nee and D’Esposito 2016; Badre and Nee
2018). Perhaps relatedly, the rostrolateral PFC has also been implicated in
tasks requiring information from memory, future directed thought, counter-
factual or alternative courses of action, or pending actions to act as control
signals (reviewed in Badre and Nee 2018). For this reason, Nee and Badre
gave this zone a general label of “schematic control” to emphasize its rela-
tionship with these types of computations.
Important to the present discussion, these distinctions along the lateral PFC
are mostly encompassed within the broadly defined FP network. However,
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Brain Networks for Cognitive Control 211
brain networks can be decomposed at multiple scales. Yeo et al. (2011) applied
a clustering procedure to a functional connectome collected at rest in a large
sample of participants. While one clustering solution, termed the 7-network
parcellation, agreed with the coarse FP- and CO-network distinction, they also
identified a finer-grained 17-network parcellation that broke up the FP net-
work into more than one network (Figure 11.2b). This included, for example, a
separate network for the rostrolateral PFC as distinct from the mid-dorsolateral
PFC and from the premotor cortex.
By comparing this 17-network structure with the functional delineations
identified by Badre and D’Esposito (2007) in the fMRI study of hierarchical
control (Figure 11.2c), Choi et al. (2018) found that there was a correspon-
dence between the functional bounds associated with task-based differences in
levels of hierarchical control and distinctions within the 17-network structure.
Further, there were also effects of hierarchical control in distinct regions of the
parietal cortex and medial frontal cortex in accord with the network structure
(Choi et al. 2018). In a direct comparison, it was found that network member-
ship, rather than rostrocaudal location, best predicted the hierarchical level of
a particular voxel (Badre and D’Esposito 2007). Thus, rather than a set of areas
or a gradient going from front to back along the lateral frontal cortex, ranked
by a factor like policy abstraction, Choi et al. (2018) found that there are a
set of subnetworks within the FP network (or MD network) that are differen-
tially activated, based on complex control demands such as policy or temporal
abstraction.
A further network property of control that has been highlighted by the study
of hierarchical control is the potential importance of corticostriatal loops
in controlling interactions between separate frontal circuits (O’Reilly and
Frank 2006; Collins and Frank 2014; Chatham and Badre 2015). It is well
established that the basal ganglia form a series of loops with the frontal
cortex via the thalamus (Alexander et al. 1986; Haber 2003). In motor con-
trol, these loops are thought to support a feedback-based gating function
(Mink 1996). Specifically, candidate actions represented by cell populations
in premotor cortex are initially too weak to fire, because thalamic drive is
under tonic inhibition by the globus pallidus. However, these candidate ac-
tions in premotor cortex also send descending inputs to the striatum. The
striatum, including putamen and caudate, receive broad inputs, not just from
this premotor region but from cortex more broadly. Cells in the striatum are
modulated by the presence of dopamine, which also induces plasticity so
that these cells can learn which combinations of actions and context have
been adaptive or not. Thus, the value of the actions considered in premotor
cortex is computed as a function of what is being processed in cortex more
broadly. If these actions have a history of being adaptive in this context,
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
212 D. Badre
“Go” cells in the striatum will elicit a cascade that ultimately disinhibits the
thalamus and allows the action to be output (Mink 1996; Wickens, 1993;
O’Reilly and Frank 2006).
One influential hypothesis is that these same corticostriatal feedback
loops can operate over goal and context representations that are needed for
cognitive control and that are maintained in working memory by the lateral
PFC (O’Reilly and Frank 2006). This computation is described using the
metaphor of a gate on working memory. When the gate is closed, informa-
tion does not pass in or out of working memory. When it is open, working
memory can be updated and top-down control signals deployed. The feed-
back loops of the basal ganglia could operate as these gates by controlling
transmission from one cortical network to another through their disinhibitory
action on the thalamus.
Consistent with this hypothesis, there is evidence from fMRI, patient, and phar-
macology studies for these corticostriatal interactions during tasks that specifically
manipulate input and output gating of working memory (Frank and O’Reilly 2006;
McNab and Klingberg 2008; Baier et al. 2010; Chatham et al. 2014; Chatham and
Badre 2015). Furthermore, the loops between the lateral PFC and the basal gan-
glia are ordered and topographic, such that there are both macro- and microlevel
loops between cortex and striatum that are arrayed in an orderly fashion along
the rostrocaudal dimension of the frontal lobes (Verstynen et al. 2012). Choi et
al. (2018) reported convergent evidence of hierarchical ordering within the stria-
tum in resting-state functional connectivity. Further, some evidence from fMRI
and TMS provides functional support for separate loops that control context- and
motor-level processing during rule learning and execution (Badre and Frank 2012;
Jeon et al. 2014; Korb et al. 2017).
Interaction among multiple corticostriatal loops is a candidate mechanism
for hierarchical control (see Figure 11.3; Frank and Badre 2012). Specifically,
the gated output of superordinate contexts maintained in working memory by
one network can act as a top-down influence on the corticostriatal gating loop
controlling subordinate networks. In this way, multiple contingent contexts can
interact hierarchically to control action.
Models of these multiple corticostriatal loop interactions have shown
that they can efficiently learn abstract hierarchical rules, transfer these struc-
tures to new tasks, and exhibit the same quasi-parallel decision dynamics
that humans employ when they perform hierarchical control tasks (Frank
and Badre 2012; Collins and Frank 2013; Ranti et al. 2015). Further, gat-
ing of contextual representations is a means of controlling input and output
through lateral PFC, thus breaking down hard problems into more manage-
able chunks. In this sense, these gating computations resemble Duncan’s
conception of an attentional episode (Duncan 2013). These computations
emerge, however, from an interaction among separate, hierarchically or-
dered subnetworks.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Brain Networks for Cognitive Control 213
GP GP GP
CN CN P
Figure 11.3 Schematic of a nested, interacting corticostriatal loop network for hierar-
chical control. The details of the corticobasal ganglia loops have been simplified in this
diagram. Each loop is a feedback loop for one cortical network. However, the output of
each network can act as a top-down influence on a lower-order loop. This nesting can
provide a mechanism for multiple-contingent gating needed for complex, hierarchical
control of behavior. Labeled areas are motor cortex (motor), dorsal premotor cortex
(PMd), anterior dorsal premotor cortex (pre-PMd), mid-dorsolateral prefrontal cortex
(mid-dlPFC), globus pallidus (GP), putamen (P), and caudate nucleus (CN). Reprinted
with permission from Badre and Nee (2018).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
214 D. Badre
drug addiction (Dalley and Robbins 2017). Not all impulsive behavior, how-
ever, is linked to inhibition tested by the SST.
Strong evidence from multiple sources has associated stopping in the SST
with a brain network that includes the right inferior frontal cortex, the pre-
supplementary motor area (preSMA), and the subthalamic nucleus (STN) (see
Figure 11.4; Aron et al. 2004, 2007; Aron and Poldrack 2006). These regions
are consistently activated in fMRI studies that employ the SST. Damage to the
right ventrolateral PFC and preSMA causes deficits in stopping that are dis-
sociable from other frontal regions, such as the dorsolateral PFC. Importantly,
the right ventrolateral PFC, preSMA, and STN interact as a dynamic network
to inhibit behavior (Aron et al. 2016; Wessel and Aron 2017). These regions
are connected by direct white matter connections, the integrity of which cor-
relates with the speed of stopping (Forstmann et al. 2012).
STN is a key node in this stopping network (Isoda and Hikosaka 2008; Li et
al. 2008; Schmidt et al. 2013). It projects an excitatory influence onto the glo-
bus pallidus, thereby enhancing its inhibitory influence over the thalamus. This
pathway can rapidly bypass the gating computations occurring in the cortico-
striatal loops and put the brakes on behavior. Recent evidence from an elegant
optogenetic study in the mouse confirms these basic features in the context of
the stopping that occurs during surprise (Fife et al. 2017). Specifically, excit-
atory stimulation of the STN cells that project to the globus pallidus caused
cessation of licking responses in a mouse. Then, inhibition of the STN elimi-
nated stopping due to a surprising stimulus.
The stopping network lies clearly distinct from the FP network involved in
contextual control that was discussed above (Aron et al. 2015). Even subcor-
tically, it appears most related to the distinct hyperdirect (rather than direct/
indirect) pathways through the basal ganglia. Thus, motor inhibition may be
another example of a dissociable form of control.
Further, there is growing evidence for a broader inhibitory role for this net-
work beyond countermanding motor actions. For example, we observed in-
creased theta band oscillations between preSMA and STN under conditions
of greater uncertainty, and this coupling correlated with slowing of responses
during the decision (Frank et al. 2015). Ostensibly through motor inhibition,
the impact of control was functionally at the level of decision making. By
stopping the output of a response, more evidence was allowed to accumulate
before committing to a response; this is formally equivalent to setting a higher
evidence threshold and making a more conservative decision. Finally, there
is evidence that components of the stopping network, including the right in-
ferior frontal cortex, may also inhibit cognitive actions, specifically the act of
retrieval from long-term memory (Guo et al. 2018; Castiglione et al. 2019). In
sum, there is a separate brain network for fast stopping, and there may also be
further subnetwork distinctions within this domain.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Brain Networks for Cognitive Control 215
M1 R-IFG
preSMA
ȕ
IFC STN
Motor
thalamus
GPi
Figure 11.4 Networks critical for stopping. (a) Cortically, the right inferior frontal
cortex (IFC), sometimes termed ventrolateral prefrontal cortex, and the presupplemen-
tary motor area (preSMA) have been consistently implicated as playing a causal role in
stopping during the Stop-Signal task (after Aron et al. 2007). (b) Schematic of the path-
ways between cortex, the subthalamic nucleus (STN), internal globus pallidum (GPi),
and thalamus that are thought to support fast stopping (after Aron et al. 2016). R-IFG:
right inferior frontal gyrus.
The evidence presented above supports either one or several networks involved
in control. However, a third perspective, most recently argued by Eisenreich et
al. (2017), holds that none of these networks truly supports cognitive control
as a unique function. Rather, since neurons are systems of distributed com-
putation, they have emergent features of control that arise naturally in such
systems. From this perspective, cognitive control is an emergent property of
network computation, and there is no specific system devoted to cognitive con-
trol in the brain (Eisenreich et al. 2017).
There are many examples of distributed systems in nature that display con-
trolled behavior without the presence of a central controller. Eisenreich et al.
(2017) gives the example of a bee swarm searching for a good site to build a
hive. Bees use dances to communicate to other bees that they have found, for
instance, a good hive site. More bees will come to the site and do the dance
if they agree with the location. Once enough bees are dancing at the site, the
dance changes to a “build here” decision. At that point, a decision threshold
has been passed, and the bees start to build. However, if there are multiple
sites, there is conflict. The bee swarms at each location grow more slowly, and
so more time is required to reach a decision. Importantly, this control adjust-
ment is carried out at a “swarm level,” not at the level of any individual bee, as
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
216 D. Badre
To some degree, merely recognizing that a network rather than a specific brain
area is important for a function like cognitive control commits the same shal-
low theoretical error as “blobology” did in the early days of cognitive neuro-
science. Labeling a network merely assigns it a location without providing
mechanistic insight or constraint on theory. However, the focus on networks
for control, rather than individual areas, does offer an opportunity to consider
new questions about the macro-level processing dynamics and functional
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Brain Networks for Cognitive Control 217
organization of those networks. Note that this is a distinct question from that
considered above, which is concerned simply with whether there are multiple
control-related networks and what their function might be. The question con-
sidered here is what the nature of the interaction among these networks, and
others, might be. In this respect, one question to emerge out of the study of
networks for cognitive control is whether these networks are hub-like or hier-
archical in organization.
Viewing cognitive control systems as hubs is an intuitive and appealing the-
oretical idea. In essence, cognitive control systems manage or modulate rout-
ing between other systems of perception and action to carry out tasks. Thus,
these networks are central to the network dynamics of the brain, will be active
across most tasks, and will exert broad influence. In other words, cognitive
control networks are flexible hubs, with near proximity to all other networks,
and with the ability to change their connectivity with multiple other systems as
needed to coordinate their dynamics during a task (Figure 11.5a).
Evidence from fMRI functional connectivity has provided some support
for the hub hypothesis. Cole et al. (2013) scanned participants while they per-
formed sixty-four different mini-tasks in the scanner. This procedure allowed
changes in connectivity to be assessed while people were shifting the rules and
domains over which they performed the tasks. Cole et al. observed that the FP
network showed the greatest variability in its connections with other networks
across all the tasks relative to any other network, including the CO network.
Furthermore, rather than just reflecting random variability in a small set of
connections, FP also had the highest participation coefficient, which derives
from how uniform its connections are across all networks. From these obser-
vations, Cole et al. (2013) concluded the FP network was acting as a flexible
hub, changing its connectivity based on the task and thereby modulating the
relevant network for a particular task. In subsequent work, this global cross-
task connectivity of the FP network has been associated with fluid intelligence,
a further clue to its potential importance for cognitive control, particularly dur-
ing rapid task instruction and execution (Cole et al. 2015).
To some degree, the flexible hub model resembles a unitary central con-
troller that is required to modulate all other dynamics in the brain. As already
noted, however, there are likely distinctions among networks for control, even
within the FP system itself. Indeed, a recent analysis of functional connectiv-
ity patterns of the FP network across multiple task conditions found that this
hub-like network was decomposable into at least two networks with different
patterns of connectivity, and that these patterns were similar to those identified
by Yeo et al.’s (2011) 17-network parcellation (Dixon et al. 2018).
An important alternative hypothesis to a global hub is that the subnetworks
for control relate to each other hierarchically, such that some networks ex-
ert higher-order influence over other subnetworks, which in turn exert control
over more restricted domains (Figure 11.5b). The cascade model (Koechlin
et al. 2003) essentially proposed such a dynamic along the rostrocaudal axis
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
218 D. Badre
(a) (c)
Sensorimotor
control
Contextual
control FEF
Schematic pre
PMd
control pre- Motor
motor
pre
Mid- PMv
(b) Hierarchical control dlPFC
IFJ
rlPFC
Figure 11.5 Hub versus hierarchical network organizations. (a) A hub network or-
ganization places a control network, like the frontoparietal network, at the center of
coordinating other networks where it serves a general and fundamental role in orga-
nizing all other networks. (b) A hierarchical network organization allows for multiple
controlling networks to share asymmetric influences with each other and to have dif-
ferences in their domains of control and proximity to other networks. (c) A schematic
summary of the results from Nee and D’Esposito (2016, 2017) showing hierarchical
interactions among frontal lobe networks (after Badre and Nee 2018). Regions along
lateral prefrontal cortex are shown within the three control zones referenced in Figure
11.2a. Heavy, unbroken arrows show strong directions of influence. Broken arrows
depict weak influences. Colored arrows are domain- or task-specific influences. Abbre-
viations: mid-dorsolateral prefrontal cortex (mid-dlPFC), rostrolateral prefrontal cortex
(rlPFC), ventral premotor cortex (prePMv), inferior frontal junction (IFJ) area, anterior
dorsal premotor cortex (pre-PMd), frontal eye field (FEF).
of the frontal lobe, such that abstract temporally extended control signals in
rostral frontal cortex influence more temporally proximate contextual signals
in lateral PFC, which in turn influence action control by premotor and motor
cortex. Other models of hierarchical control have shared similar dynamics,
including among nested corticostriatal loops and through medial PFC (Frank
and Badre 2012; Alexander and Brown 2015).
In a set of two fMRI experiments, Nee and D’Esposito (2016, 2017) pro-
vided evidence for a hierarchical structure within lateral PFC. These studies
used estimates of effective or directional connectivity from dynamic causal
modeling while subjects performed a set of complex tasks that engaged vary-
ing degrees of hierarchical control in verbal versus spatial input domains.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Brain Networks for Cognitive Control 219
Hierarchical strength was defined in terms of greater outward than inward con-
nectivity (i.e., a region has broader outputs than inputs, as defined in Badre et
al. 2009).
The basic results from these experiments are summarized in Figure 11.5c:
Mid-dorsolateral PFC was active in higher- (more abstract) but not lower-
order tasks across both input domains. It exerted an influence on the more
caudal dorsal premotor cortex and ventral premotor cortex regions that were
active across both the simpler and more complex tasks, but only in the spatial
or verbal domain, respectively. These caudal contextual control regions also
received domain-specific input from sensorimotor regions. The mid-dorsolat-
eral PFC received greater input from the rostrolateral PFC during conditions
where temporal abstraction was required. In a follow-up TMS study, Nee and
D’Esposito replicated these findings and showed that stimulation of nodes in
this network produced behavioral effects that were broadly consistent with this
information flow.
These findings from fMRI in humans converge with earlier anatomical stud-
ies in the macaque monkey. Goulas et al. (2014) performed an extensive meta-
analysis of monkey anatomical projections using the CoCoMac database and
focused on the connectional asymmetry that might drive hierarchy. They coded
multiple sites in the PFC based on the same definition of hierarchy as above:
any area higher in the hierarchy would have broader efferent connections to
lower-order areas than the reverse. Consistent with Nee and D’Esposito, anterior
mid-dorsolateral PFC (areas 45 and 46) showed the greatest asymmetry on this
metric, relative to regions caudal to the mid-dorsolateral PFC or to the rostrolat-
eral PFC which is anterior to it (Goulas et al. 2014). Notably, although Goulas et
al. (2014) did find evidence that the mid-dorsolateral PFC was higher in terms
of this network definition of hierarchy, it was not the most hub-like, based on a
measure of betweenness centrality. This appears consistent with structural con-
nectivity metrics in humans as well (van den Heuvel and Sporns 2013).
It remains open how one should characterize the dynamics among networks
supporting cognitive control. The broadly defined FP system exhibits a hub-like
character, with high participation and flexibility in connectivity across multiple
tasks. There is also evidence that subnetworks within this overall system re-
late to each other hierarchically. In that system, there is no central domain
general hub. Rather, the rostral mid-dorsolateral PFC is not active or neces-
sary across all tasks; it is necessary during those complex tasks that require
higher-order contextual control. Lower-order areas within the FP system are
activated across more tasks, but they are domain specific. Thus, a hierarchical
control architecture assumes that global control of the whole system emerges
from limited, local, and hierarchical interactions among control networks. This
contrasts with a hub network that manages interactions broadly and globally.
Finally, it should be noted that there are other hierarchical models we
have not discussed that yield different organizations. For example, Barbas
and Rempel-Clower (1997) proposed a laminar definition of hierarchy which
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
220 D. Badre
distinguishes regions based on their output versus input layers of cortex. This
laminar definition of hierarchy also predicts hierarchical interactions within
the FP system, but places the more rostral areas, like rostrolateral PFC, higher
(Goulas et al. 2014). Thus, the architecture and organization of networks for
control remains a mostly open question at present, but at least one core distinc-
tion is between those proposing hub-based interactions and those proposing
hierarchical ones.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Brain Networks for Cognitive Control 221
we are using general rules. Now consider that you have a particular route that
is being used a lot. You could build an express road between those two loca-
tions in town: at some cost of time and asphalt, you then add some dimensions
to your road system and gain a low traffic route. Increasing your dimensional-
ity is costly but highly efficient, if you know you need a particular set of routes.
From this analogy, one way to think about the transition from controlled to
automatic behavior is to view it as a transition from a reliance on generaliz-
able, low-dimensional neural representations that are subject to interference
to high-dimensional neural representations that take time to build but which
directly map a combination of inputs to a response. These transformations
could occur as transformations within frontal systems themselves. Early on,
coordination among more networks, using gates and so forth, is necessary
because a new task has to be assembled from low-dimensional components.
However, over time, it is efficiently supported by a high-dimensional repre-
sentation that allows a more direct route from input to output. It is still routed
through the PFC, where multiple contexts and goals can affect it, but just
differently in terms of the format of the routing (e.g., from low to high dimen-
sions). This is different from the modulatory view which requires that there
are always the same separate tracks from input to output: control acts like a
switch operator deciding which track gets to run and when. This is among
the distinctions that Eisenreich et al. (2017) made in their argument about
emergent control systems and is captured by their schematic representations
of different control architectures (Figure 11.6).
The evidence for this transmissive rather than modulatory model of control
is limited at present, but there are intriguing clues. First, the computational
trade-off described above between generalizable low-dimensional representa-
tions versus parallel high-dimensional representations has been shown in theo-
retical work using neural networks (Fusi et al. 2016; Musslick et al. 2017).
Second, there is evidence from physiology in the nonhuman primate and mul-
tivoxel pattern analyses of human fMRI data that the FP network does not
encode single contextual features of tasks but rather large conjunctions of mul-
tiple task features (Woolgar et al. 2011, 2016; Rigotti et al. 2013; Pischedda et
al. 2017). Presently, we lack clear evidence that separate areas or networks rep-
resent separate contexts or elements of a task. Third, maintenance in working
memory may not be a fixed-point system, wherein information is maintained
in a single stable form to be accessed at any point as an external control sig-
nal. Rather, evidence from electrophysiology in monkeys and EEG in humans
suggests that neural ensembles undergo dynamic change over time (Stokes et
al. 2013). Thus, these representations are themselves expressed in trajectories
toward an end point. Finally, evidence from nonhuman primates has shown
that the nonlinear mixed selectivity of PFC neural representations supports
high-dimensional capacity during task performance (Rigotti et al. 2013). This
is what allows these populations to encode multiple mixtures of their inputs
in unique patterns that can be read out by downstream cells. Rigotti et al. also
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
222 D. Badre
Modulatory control
Controller
network
Processor
Stimulus A Action A
Stimulus B Action B
Transmissive control
Controller/processor
network
Stimulus A Action A
Stimulus B Action B
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Brain Networks for Cognitive Control 223
The conception of the mind and brain as a control system is one of executive
function’s most animating theoretical ideas, dating back at least to Norbert
Wiener’s mental-servo notions in Cybernetics (Wiener 1948) and the semi-
nal studies on the human operator in motor control by Kenneth Craik (1948).
These ideas and their descendants rely on the engineering formalism of opti-
mal control theory. In a control system, there is a set point, which is the desired
system state, and mechanisms of feedback or prediction that lead the system to
adjust toward the set point either in response to or in anticipation of changes
to the system state.
In the classic example of a thermostat, the state of the system that mat-
ters is the temperature in your home. The set point is the desired temperature.
Feedback to the system in the form of temperature measurements can result in
heating or cooling actions that will change the temperature of the environment
until the set point is reached. This is feedback control. Fancier modern ther-
mostat systems may also anticipate or learn about how ambient temperatures
change over the course of a day. Such a system can engage proactive cooling or
heating to maintain a stable set point, and so implement feedforward control.
Regardless of its specifics, however, the efficiency with which a control system
can reach its set point and the range of set points it can reach are a means of as-
sessing its quality. Control systems can be evaluated, compared, and optimized
on the basis of their ability to reach any particular desired state from any initial
state (termed controllability) and the efficiency with which they do so.
In cognitive control, control theory concepts have been historically posed
at the cognitive-functional level. The set point is defined with reference to
some real-world defined goal or target, such as drinking coffee or making it to
your connecting flight or naming the ink color in a Stroop task. An effective
control system is one that allows us to reach the widest range of such goal
states efficiently, either in the world or our cognitive system, given a similarly
wide range of initial contexts and situations. In this conception, maximal con-
trollability (i.e., being able to get to any output state given any input state) is
presumably what cognitive neuroscientists intuitively mean when they use the
term flexible behavior.
From a control theory perspective, psychological or neural mechanisms
must gather feedback or make predictions about the distance to desired set
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
224 D. Badre
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Brain Networks for Cognitive Control 225
0 234 0 234
Average controllability Modal controllability
1% 3% 3%
12% 2% 16%
11% 17%
11%
19%
15%
30% 7%
4% 1%
12%
18% 18%
set point, then, is a target pattern of brain activity, not a real-world objective,
and the control system must plot the distance between your starting pattern of
brain activity and that goal pattern of brain activity through a functional con-
nectome. Finally, controllability is not defined in terms of how you behave but
rather how readily you can shift from the brain state you are in to any desired
brain state.
So, why aren’t these levels of theorizing about control the same? Isn’t this
network conception just a reductionist reframing of the original functional-
level control problem in terms of brain states? This is certainly the way it is
often posed and interpreted. However, a complication arises because of the
classic philosophical problem of multiple realizability. This is exactly the case
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
226 D. Badre
where this abstract philosopher’s thought problem has some real implications
for scientific theory.
Multiple realizability was famously raised by Hilary Putnam as an argu-
ment against identity theories in which each psychological state had one and
only one implementation (Putnam 1967; Block and Fodor 1972). The argu-
ment is that multiple species or individuals within a species can realize the
same psychological process, like pain or vision, with very different brains.
Thus, there is a many-to-one mapping of brain states to psychological states.
David Marr’s idea of distinct levels of analysis relies on similar arguments
about the asymmetry as one goes from computational to algorithmic to imple-
mentational levels (Marr 1982). Essentially, the functional or psychological
level is abstraction over multiple possible realizations of algorithms and imple-
mentations within the brain.
The implications of this concept for cognitive control are important. For the
reductionist reframing to correspond directly to the functional-level models,
there must be a one-to-one correspondence between any one functional-level
state (e.g., drinking) and a corresponding brain state (or a highly correlated
class of such states). However, as the problem of multiple realizability high-
lights, this assumption is hardly guaranteed. Rather, the goal of having that
drink may entail a wide range of activities to get there and a wide set of pos-
sible realizations of actually quenching one’s thirst. Each of these is associated
with a set of brain states. Some may not be more strongly correlated with
each other than with other states, and so comprise a disjunct set. Thus, what
ultimately connects this disjunct class of brain states is the functional-level
outcome, drinking. As such, conducting control at the functional level would
be the best way to ensure success, rather than making specific brain states a
set point.
A further issue concerns feedback in a brain state control system. To work
at the brain state level, the control system needs a means of detecting its dis-
tance from its set point. But, we don’t have explicit access through the senses
into our actual brain state, in terms of what neurons are firing and when. We
do, however, have access to a functional-level description, like whether we are
drinking or not. This feedback is also essential for the control system to learn
and know what actions to take to reach a goal in the future. Without assuming
a direct correspondence between the functional and brain state level that vio-
lates multiple realizability, feedback about the specific brain state target is not
available to the control system in an obvious way. This is a problem if one’s
control system is operating primarily at a brain state level. We do, however,
have perceptual systems that can assess the real-world outcome of our actions
and can use these to assess our state.
What this discussion highlights is that computing the distance to a goal
and seeking inputs that minimize that distance is different, depending on
whether one is mapping the distance to a particular brain state or the distance
to a disjoint set of such states defined by a functional outcome. Even if one
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Brain Networks for Cognitive Control 227
allows for some correlation among that set of states, it is easy to see how the
problem gets quite complicated in the latter case, if the control system has no
access to the functional-level description and is instead optimizing control
over brain states.
These complications notwithstanding, the issue of level of controllability
remains unresolved. It is not clear to me that the conventional functional view
is necessarily always the correct one or whether both options might not be
true and influence matters under different circumstances. At the level of plan-
ning, awareness, and explicit control, the functional-level description of con-
trol might be the appropriate level at which to understand cognitive control
for the reasons discussed above. As such, learning and feedback mechanisms
must ultimately reference this level of analysis. Neural accounts must explain
how the brain supports this functional-level control system, deploying neural
mechanisms that interpret the state of the world with respect to goals, compute
distances to real-world hypothetical and counterfactual outcomes, define the
means to cross that space, and monitor progress as it goes. However, for other
kinds of control, such as switching among well-learned tasks or adjusting on
the fly to maintain a stable trajectory of behavior, things may be different. In
these cases, the principles and constraints of brain-level network control may
be the most relevant feature determining individual variability in success, even
for functional-level outcomes.
In sum, whether control plays out at a primarily functional versus brain
level is an important open question, particularly as we take more sophisticated
approaches to understanding dynamics within the brain’s connectome. These
levels of controllability are not mutually exclusive, as both may influence con-
trolled behavior. This discussion highlights the importance of being explicit
about the level at which we assume control is occurring in our theorizing,
and that it is not trivial to assume that brain state control is isomorphic with
functional-level control.
Concluding Thoughts
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
228 D. Badre
Acknowledgments
This work was supported by the National Institute of Mental Health (MH111737) and
National Institute of Neurological Disease and Stroke (NS108380) of the NIH, a MURI
award from the Office of Naval Research (N00014-16-1-2832), and an award from the
James S. McDonnell Foundation. I am grateful to Apoorva Bhandari, Olga Lositsky,
and other Badre Lab members for helpful comments and discussions on these topics,
and to Emily Chicklis for her help with figures and illustrations. Figure 11.7, from Gu
et al. (2015), and Figure 11.4b, from Aron et al. (2016), are used under the terms of the
Creative Commons Attribution 4.0 International License (CC-BY). Copyright for Fig-
ure 11.4a, from Aron et al. (2007), is held by the Society for Neuroscience
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
12
A Framework for
Understanding Agency
Kayuet Liu and Hakwan Lau
Abstract
Why do some thoughts feel involuntary and intrusive? When should we hold someone
responsible for their actions and thoughts when they all have some basis in the brain? Are
we truly free agents when we are bounded by shared values and culture? This chapter
presents a framework for how our consciousness of our own intentions and emotions
allows us to form causal narratives about ourselves and the world. These narratives de-
termine our sense of agency, and we ascribe responsibility correctly depending on the ex-
tent to which one is capable of forming culturally appropriate narratives. Different ways
of characterizing consciousness are analyzed, with a focus on one that may prove most
useful within the context of understanding individual agency. A variant of the higher-
order view of consciousness is advocated that allows us to form causal, albeit imperfect,
narratives about ourselves. However, it is because of these imperfect narratives that our
understanding of agency and responsibility is formed. Thus, understanding how these
narratives come about is an important first step to understanding agency and how some
thoughts are considered involuntary and intrusive. Implications of this framework are
discussed using examples from mental illnesses, addiction, suicide, and racism.
Aberrant acts committed by patients with severe mental illnesses (e.g., schizo-
phrenia) are often considered not punishable. By law, if patients have lost
their capacity to reason, society should not hold them criminally responsible
(Mobbs et al. 2007). However, there seems to be a spectrum of controllability
(Moscarello and Hartley 2017) within which we ascribe degrees of respon-
sibility to patients suffering from different mental disorders. Take addiction
as a contrasting example. Law aside, members of society often disagree on
whether addicts are responsible for their own actions. Some hold that it is the
addicts’ own “decision” to go down the path of becoming who they are. Some
have challenged the notion that addiction is a brain disease, because the neural
correlates of addiction are not sufficient to cause addicts to do what they do in
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
230 K. Liu and H. Lau
many environments (Levy 2013). The question, then, is: How “voluntary” are
the behaviors of addicts (e.g., seeking out a drug) compared to those of patients
with schizophrenia (e.g., talking to an imaginary friend in public), since both
have bases in the brain? Indeed, when a healthy student is late for class due
to procrastination rather than a traffic jam, this behavior is also based in the
brain. In a sense, people’s brains are the causes of their behaviors in all these
instances. Yet we ascribe agency and responsibility differently in each case.
Perhaps this relates to how we see an individual’s agency and responsibility
in the context of culture. If someone is brought up in a society in which it is
acceptable to take food off each other’s plates without first asking, much as we
may disapprove of such actions, we may accept that person’s behavior more
easily than had that person been brought up in a culture where such behavior
was sternly forbidden. In this sense, are we truly free individuals, acting volun-
tarily out of our own desires or judgments? Or are we bounded very much by
our shared values, so that our errors may reflect more on the failure of society
rather than ourselves?
In this chapter, we do not attempt to solve these difficult questions, but
rather hope to provide a useful framework within which they can be addressed.
We will argue that the notion of consciousness is crucial to understanding these
issues. However, there are many different ways to characterize consciousness.
We evaluate a few accounts and focus on one that may prove useful within the
context of understanding individual agency.
Traditionally, debates on free will concern whether our actions are predeter-
mined; that is, whether our actions are genuinely de novo. The assumption is
that, in principle, if we knew all current physical events of the world, together
with a complete understanding of all physical laws, we should be able to pre-
dict the next events perfectly (Laplace 1951; Hoefer 2016). Within this con-
text of determinism (Hoefer 2016), how can one be an “unmoved first mover”
(Strawson 1994; Pereboom 2001)? Are our actions not already fully determined
physically, before they actually take place? To the extent that some notion of
freedom is possible within the deterministic framework (i.e., compatibilism;
Fischer 2006; Nahmias 2016), it cannot be because these actions are random
and thus unpredictable. Rather, we are responsible for our actions because we
have some degree of autonomous control over our actions. To argue that our
actions are genuinely free in this sense, one option is to argue that the physical
world is not truly fully deterministic. Some have appealed to findings from
quantum physics: that certain future events are not fully determined, even if all
current physical events are known (Kane 1996). This debate is important and
interesting, but many good reviews of the literature already exist (e.g., Sinnott-
Armstrong 2008). Here we focus specifically on considering which cognitive
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
A Framework for Understanding Agency 231
Shifting from the traditional focus on whether our actions are predetermined,
it has been argued that for individuals to be held responsible for their actions,
they need to be conscious of certain relevant events (Caruso 2012; Levy 2014).
Intuitively, this makes sense: it seems unfair to hold one fully accountable for
something that one isn’t even aware of having done, nor having even remotely
contemplated doing. Incidentally, there is experimental evidence that this
“consciousness requirement” is in line with our folk psychological concepts of
free will and responsibility (Shepherd 2017). Accordingly, many have focused
on the question: To what extent do conscious mental states truly cause behavior
(Pockett et al. 2006)? This diverges from the traditional question of determin-
ism, because this new question is still meaningful even if consciousness itself
were fully deterministic (Nahmias 2014). So long as the (deterministic) con-
scious processes in the brain causally influence our behaviors, there may still
be an important sense in which we have control over our actions.
There are, of course, critics of the view that consciousness is relevant. For
example, Smith (2005) claims that forgetting a close friend’s birthday (i.e.,
something that one does not consciously choose to do) does not eliminate the
responsibility of failing to call or send a card. We will not go into the details
here (for further discussion, see Caruso 2012), but an important lesson to de-
rive from these exchanges is that the arguments often depend on which specific
notion of consciousness is at play. We will focus on this issue in the next few
sections.
The question of whether the conscious processes in the brain are causal to be-
haviors may seem trivial (Baumeister et al. 2011). Although we all feel that our
conscious thoughts and decisions have causal efficacy, several lines of empiri-
cal studies seem to challenge this intuition. First, studies of unconscious prim-
ing, mostly coming from the area of social cognition, show that our actions and
decisions may be influenced by unconscious cues (e.g., words or symbols ir-
relevant to the primary tasks at hand, or the gender or age of the experimenter),
the meaning of which we are not fully aware (Bargh et al. 2012). If true, these
findings may show that our actions are not as fully consciously determined as
we thought. Some have called into question, however, whether these findings
can be replicated (Chabris et al. 2019), and one could argue that the relevant
cues are merely unattended but not truly subliminal.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
232 K. Liu and H. Lau
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
A Framework for Understanding Agency 233
that the subject will subsequently show reduced physiological threat-related re-
sponses to images of spiders (Taschereau-Dumouchel et al. 2018a). In other stud-
ies, using a similar experimental setup, powerful forms of reward-based learning
have been shown to take place unconsciously (Cortese et al. 2019). In addition,
subjective confidence regarding one’s ability to discriminate certain stimuli can
also be changed with this unconscious association method (Cortese et al. 2016).
In summary, unconscious cognitive processing seems very powerful indeed,
especially regarding its ability to form statistical associations and to influence
subsequent behavior (via priming). Does this mean that consciousness plays
no causal role and has no function? The answer is not so straightforward. Lau
(2009) argues that although unconscious processes are powerful, this does not
mean that there is necessarily no room for consciousness to add further ben-
efits. To discuss meaningfully the theoretical possibilities of the role for con-
sciousness, we need to distinguish between different notions of consciousness.
A Middle Ground?
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
234 K. Liu and H. Lau
Just because a philosophical notion of consciousness exists and can serve our
purpose does not mean that we should adopt it. Fortunately, there is consider-
able empirical support for the higher-order view of consciousness (Lau and
Rosenthal 2011). Neuroimaging studies have shown that subjective aware-
ness of visual stimuli is associated with brain activity in high-level cognitive
regions (e.g., in the prefrontal cortex), even under highly controlled experi-
mental conditions in which the subjects are not merely processing the stimuli
in simple (first-order) tasks (Lau and Passingham 2006). Also, neurological
patients with selective damage of their visual cortex may lose the relevant
subjective visual experiences but not their ability to correctly “guess” the
identity of the relevant stimuli (Weiskrantz 1997); when visual stimuli are
presented to intact parts of their visual cortex (as compared to the “blind”
regions) leading to a conscious experience, higher activity in the prefrontal
cortex was also found (Persaud et al. 2011). Disruption of activity in this
prefrontal brain region selectively impairs one’s ability to introspect whether
one has successfully perceived the stimuli, without impairing (first-order)
perception itself or memory (Rounis et al. 2010; Fleming et al. 2014). These
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
A Framework for Understanding Agency 235
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
236 K. Liu and H. Lau
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
A Framework for Understanding Agency 237
Taking an analogy from computers, the ability to form such beliefs seems
useful. Let us say that a computer server in a large network sends a command
to print ten pages to the printer. Another node in the network complains that the
printer queue is now jammed and asks the first server to resolve the problem. It
would be quite useful for the first server to know who contributed to the print-
ing in the first place.
This self-directed nature of the corresponding representations is also rel-
evant in emotions. Again, we know that for certain basic emotions, imagining
them activates similar neural representations as experiencing them (Reddan
et al. 2018). Likewise, when we emphasize and think about others’ emotions,
similar representations are involved. Therefore, when these first-order emo-
tional states occur, the brain needs to know what the causes are. According
to the view advocated here, one consciously experiences emotions when the
relevant higher-order state points out that the first-order emotional representa-
tion reflects what one is going through. Thus, the corresponding belief may
be simply that “I am angry” or “I am scared.”2 Having such beliefs may be
quite useful in navigating social situations and in explaining to others why we
behave a certain way.
Contrasting these with the relatively simple beliefs in the case of perception
(“there is a cat”; see Figure 12.1), there is a sense that agency and emotions
are intrinsically more self-involving. In fact, as Ledoux argues, without some
minimal concept of the self, an animal may not experience basic emotions
(e.g., fear) at all (LeDoux and Sorrentino 2019).
Why does one need to form these rational beliefs about oneself, which
requires that the relevant first-order states be made conscious via meta-rep-
resentations? The proposal is that first-order processes are powerful: we can
use them to learn about statistical associations between events. However,
mere associations are not coherent narratives. Moreover, narratives are sto-
ries in which events are causally related. When we say that Julius Caesar
invaded a certain country because he was angry, we mean that his emotion
caused certain behavior. When he decided to invade, presumably he saw
himself as an agent who was causally responsible for the decision. Inferring
about causality is, however, notoriously hard. It is technically a challeng-
ing problem from a statistical point of view (Pearl and Mackenzie 2018).
Without controlled experiments in which we can manipulate the putative
causes while holding all other things constant, assumptions need to be made
and heuristics involving counterfactuals may need to be invoked (Bond et al.
2012; Chambon et al. 2018). As such, interpretations matter; there may be
more than one way to tell a story based on the same facts. With these imper-
fect narratives, we form a quasi-rational understanding of why we behave a
certain way, and we provide socially acceptable justifications of our actions,
based on folk psychology.
2
Forming such beliefs does not necessarily involve language ability.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
238 K. Liu and H. Lau
There is no
I want to
Rational beliefs There is a cat lion; it is just I’m afraid
run
my imagation
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
A Framework for Understanding Agency 239
versus model-free distinction may map onto conscious versus unconscious pro-
cesses, such mapping is not intended to be clear cut. The higher-order view, on
the other hand, is a theory of consciousness per se with direct empirical evidence.
In the present context of understanding agency, the consciousness requirement
is an important component of the overall argument. Accordingly, it is not clear if
one should be absolved of the relevant responsibility just because decisions are
made with System 1 (fast) thinking or model-free learning.
Another distinction is that one may conceive of the Two Systems framework
as representing two parallel processes. The first- versus higher-order model
here, however, stipulates that the two mechanisms are in a hierarchy. This hi-
erarchical nature may have important consequences for treatment of mental
disorders; intervention at the first level will causally impact on the higher level
(Taschereau-Dumouchel et al. 2018b).
This self-narrative account of agency may help us understand why some be-
lieve that patients suffering from severe mental illnesses may be less deserving
of punishment than the unpunctual student, even though in both cases brain ac-
tivity causes the relevant behavior. In the case of a patient with schizophrenia,
the very basic mechanism of higher-order perceptual reality monitoring may
be at fault. Thus, the patient may be unable to distinguish self-generated inner
speech from externally triggered voices (e.g., from “God”), occurring from a
breakdown between the bottom two levels in Figure 12.1. The patient may not
be able to tell if an action is voluntarily produced or controlled by aliens. As
such, the entire self-narrative system may well disintegrate. It is probably not
fair to hold such patients accountable for their own behavior, if they are not
correctly aware of who and what events caused these actions in the first place.
In the case of the unpunctual student, the higher-order system is presum-
ably intact. What might have caused the (mildly) delinquent behavior may be a
momentary overemphasis on the value of not having to rush or an attraction to
another activity. These values represented in the first-order system are no less
brain based and, in a sense, they too cause the resultant behavior. With an in-
tact higher-order system, however, one should be able to appreciate that these
first-order values are problematic and that one would be guilty all the same for
acting a certain way.
What about cases of addiction and substance abuse? Between severe men-
tal illness and everyday cases of delinquency, there likely lies a spectrum. In
some cases of addiction, one may suffer first-order malfunctioning such that a
substance may be associated with an unrealistic expected level of immediate
reward, even when one is well aware that it cannot be good in the long run. In
some cases, this higher-order mechanism may well be relatively intact, so one
may be accountable for not recognizing the situation as such. However, there
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
240 K. Liu and H. Lau
Mental illnesses can sometimes lead to one of the most devastating conse-
quences: suicide. At times considered to be one of the most personal and
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
A Framework for Understanding Agency 241
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
242 K. Liu and H. Lau
social discourse. At times, focusing on this higher level may well be more ef-
fective than methods focusing on changing our first-order implicit biases.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
A Framework for Understanding Agency 243
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
13
Systems Approach to
Intrusive Experiences
Angela C. Roberts, Rita Z. Goldstein, David Badre,
Bernard W. Balleine, Hugo D. Critchley,
Aikaterini Fotopoulou, Sophia Frangou,
Karl J. Friston, Tiago V. Maia, and Elliot A. Stein
Introduction
This chapter explores how intrusive experiences may occur at a systems level from psy-
chological, computational, neurobiological, and physiological perspectives. A general
scheme is proposed of the essential elements of an intrusive experience, and where in
this scheme dysregulation could occur to increase the likelihood of an intrusive experi-
ence. It also considers a range of psychological and mathematical models that have
been applied to explain how intrusions may ultimately happen, some of which are more
closely integrated into neurobiological systems than others. These include a Bayes-
ian model of active inference, integrated psychological and physiological models of
interoception, and psychological and neurobiological models of working memory and
associative learning and their relevance to concepts of flexibility and stability.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
246 A. C. Roberts et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems Approach to Intrusive Experiences 247
The Intrusion
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
248
IC
IC
dACC
dACC
Amy
TP
Integration
monitoring
systems
A Model of the Brain
cognition system
cognitive system
Goal-directed
Spontaneous
Homeostatic
Sensory
systems
systems
systems
Motor
Hypo
Medulla
PAG
LCC
NTS
Amy
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Figure 13.2 The brain is functionally organized into cognitive systems supported by spatially defined networks (Power et al. 2011). This simplified
model illustrates brain activity within brain systems (left) that could host a potential “locus of origin” of an intrusion: sensory and motor systems,
homeostatic systems involved in the integration of exteroceptive and interoceptive signals (Salvato et al. 2020), a system involved in spontaneous
cognition supported by the default mode network (Raichle et al. 2001), and a system for goal-directed behavior supported by frontoparietal regions
(Fox et al. 2005). The model posits that intrusions are experienced as such when they enter “awareness,” which is likely to occur in the presence
of additional recruitment of networks involved in monitoring (shown on the right), such as the salience network (Seeley et al. 2007). The blue ar-
eas (middle) represent the network space, with oscillatory activity therein denoted by red lines. Amy (amygdala), dACC (dorsal anterior cingulate
cortex), Hypo (hypothalamus), IC (insula cortex), PAG (periaqueductal gray), LCC (lateral cerebral cortex), NTS (nucleus tractus solitarius), TP
(temporal pole).
Increased strength
Normal threshold
Reduced threshold
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
250 A. C. Roberts et al.
discuss in more detail below). Here we use the analogy of awareness thresh-
old (borrowed from sensory perception) to visualize the moment a signal
gains sufficient biological momentum to breach the threshold of awareness.
Accordingly, breach intrusive experiences can occur because the strength,
features, or contextual significance of the originating signal enables its selec-
tive enhancement. In contrast, permissive intrusive experiences occur when
the threshold is transiently or persistently lowered and thus permits the propa-
gation of weaker signals. The timing of intrusive experiences (i.e., when they
occur) can be influenced at any point by the external environment as well as
by internal states, which can be referred to as “motivational states” in that
they combine representations of somatic states and overall general behavioral
drives (discussed in more detail below). It also follows that intrusive experi-
ences are influenced by genetic and molecular factors, including neurotrans-
mitters (e.g., Bonvicini et al. 2016; Sinopoli et al. 2017), that define healthy
within-individual variation (i.e., the likelihood of an intrusion within an indi-
vidual) and interindividual differences (i.e., differences between individuals
in the likelihood of experiencing intrusions), and that these may be associated
with pathological conditions affecting brain integrity at multiple organiza-
tional levels (e.g., Keelan et al. 2019).
Generally, signals relating to survival (e.g., hypoglycemia) will generate
breach intrusive experiences. The same could apply to abnormally generated
signals, as in the case of Tourette syndrome, where abnormal sensorimotor ac-
tivation spreads to other brain regions (e.g., the insula) and eventually breaches
the threshold of awareness (Conceição et al. 2017). Signals relating to signifi-
cant prior (e.g., childhood abuse, traumatic event) or immediate circumstances
(e.g., negative thoughts about the self) may also be selectively enhanced and
thus breach the awareness threshold. In such cases, the content of the intrusive
experiences is more likely to be “personal” to the individual. The personal
nature of the intrusion is also likely to constrain the range of its content; thus,
such intrusive experiences are likely to be stereotypical. The intrusion experi-
ences observed in posttraumatic stress disorder (PTSD) are prime examples as
their content is repetitious and of personal significance (American Psychiatric
Association 2013). Our model also predicts that permissive intrusive experi-
ences are likely to have a more variable and circumstantial content because the
lowering of the awareness threshold will permit the propagation of a variety of
signals. Attention deficit hyperactivity disorder (ADHD) would be a prototypi-
cal example of a condition in which permissive intrusive events might occur.
Currently, intrusive experiences in ADHD are considered in terms of abnor-
malities in attentional brain systems that gate awareness (Castellanos and Proal
2012; Bozhilova et al. 2018). As already mentioned, the dichotomization of
intrusive experiences as breach or permissive does not imply that they are mu-
tually exclusive. For example, up to 50% of patients with Tourette syndrome
have ADHD, suggesting that breach and permissive intrusions may co-occur
and determine clinical severity and complexity.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems Approach to Intrusive Experiences 251
The Appraisal
During the appraisal stage, we postulate that the intrusive experience will at-
tract unconditional and conditional attributes and emotional states. By defini-
tion, intrusive experiences will be unconditionally labeled as involuntary as
they bypass processes of agency (see Liu and Lau, this volume; Gallagher
2012; Moore and Fletcher 2012; Braun et al. 2018). However, typical intrusive
experiences retain the “sense of ownership”; that is, the sense of selfhood we
attribute to our own bodily sensations, thoughts, and actions (Gallagher 2012).
It is worth noting that they are distinct from psychotic experiences which, al-
though often construed as intrusive (especially hallucinations and delusions),
typically involve a loss of agency and self-ownership (Feinberg 1978; Moore
and Fletcher 2012; Frith 2014).
The appraisal of intrusive experiences is a multisystem phenomenon that
may, in some cases, rely on complex representations involving semantic/
linguistic networks. During appraisal, the attributes assigned to intrusive ex-
periences and the emotional responses they invoke will depend on their con-
tent, nature, and normative significance (i.e., alignment of personal beliefs
and societal values) (Korsgaard 2009). We argue that the ultimate purpose of
the appraisal is to determine the “likedness” of the intrusive experience; that
is, the degree to which the experience is aligned with the individual’s future
plans (Figure 13.4). As used here, likedness aligns with notions of motivational
relevance (Higgins 2011) and self-congruence (Rogers 1959; Higgins 1987)
and, as mentioned above, the appraisal of the intrusive experience depends
on the characteristics of the individual having the experience, including their
exposures.
Appraisal
Liked Unliked
Help
seeking
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
252 A. C. Roberts et al.
Intrusive experiences that are appraised as distressing and “not liked” are
more likely to be classified as clinically significant and to elicit help-seeking
behavior. However, intrusive experiences can be of a positive nature and
liked, as in thoughts associated with loved ones or that emerge from sud-
den insight or “eureka” moments (Kounios and Beeman 2014). Still, intru-
sive experiences that are deemed positive are not always adaptive and may
contribute to further pathology by providing confirmation for maladaptive
beliefs, as in hedonic hunger in individuals with restrictive eating disorders
(Lowe et al. 2016).
The Outcome
The outcome of the appraisal will invoke mechanisms and networks that sup-
port selective attention, decision making, response inhibition, and response
selection (Niendam et al. 2012; Langner and Eickhoff 2013; Zhang et al. 2017;
Chen et al. 2018b). We assume that there will be no voluntary inhibition for
liked intrusive experiences (Figure 13.5). The experience would either be al-
lowed to decay or it could be maintained through attentional mechanisms. A
liked intrusive experience may even act as a catalyst or starting point for an-
other mental or motor plan. In such cases, the switch from the pre-intrusion
state to a new one may be viewed as a positive outcome of the intrusive events.
Eureka moments would fall under this category.
By contrast, “unliked” intrusive experiences will evoke attempts at volun-
tary inhibition. The success or failure of the experience will depend on the
functional integrity of frontostriatal networks that are generally implicated
Outcome
Liked
Unliked
Voluntary
Inhibit
perseveration
Liked Unliked
Switch to new Involuntary
process perseveration
Help
Normal
seeking
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems Approach to Intrusive Experiences 253
in inhibitory control (Niendam et al. 2012; see also chapters by Balleine and
Badre, this volume; Bari and Robbins 2013). Of note, disorders characterized
by intrusive experiences also present with a more general impairment of inhib-
itory control that affects multiple aspects of cognition and behavior (Gourley
et al., this volume; Marsh et al. 2009; Shin et al. 2014; Morand-Beaulieu et al.
2017; Pievsky and McGrath 2018). Failure of inhibitory control is expected
to give rise to perseveration and/or premature action (as exemplified in com-
pulsivity and impulsivity, respectively), which may elicit secondary appraisals
involving frustration, anger, and increased arousal directed at the failure to
inhibit rather than the original intrusive experience. Such an outcome is likely
to increase the allocation of attentional resources to the intrusive experience
and the inhibitory failure; in some individuals, this may reinforce intrusion
experiences, leading to a pathological loop.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
254 A. C. Roberts et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems Approach to Intrusive Experiences 255
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
256 A. C. Roberts et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems Approach to Intrusive Experiences 257
continue to be in play and induce successive checking behavior that may pro-
ceed indefinitely.
Notice in this example that checking behavior has been formulated in terms
of aberrant salience because the action of rechecking the door does not lead to
the resolution of uncertainty. This aberrant salience is suboptimal (i.e., patho-
logical) because of a failure to attenuate or change interoceptive signals. In
short, a failure of sensory attenuation led to aberrant salience and a persistent
epistemic affordance that never resolves itself. In other words, no matter how
many times I check the door, I never sense that my uncertainty has been re-
solved, which could further maintain a state of autonomic arousal. Expressed
even more simply, this checking behavior is futile because there is an irre-
ducible uncertainty about the state of the world due to a failure to attenuate
interoceptive evidence from my body. This predictive processing, or active
inference account of OCD, is based (and elaborates) on work by Kiverstein et
al. (2019) and Rae et al. (2019a), and owes much to seminal accounts of why
patients with OCD appear to be “stuck in a loop.”2 For example, Roger Pitman
(1987:336) suggested that “the core problem in OCD is the persistence of high
error signals, or mismatch, that cannot be reduced to zero through behavioral
output,” and that “the obsessive-compulsive’s internal comparator mechanism
is faulty. No matter what perceptual input it receives, it continues to register
mismatch….It may be that in fact the action was well done, but the defective
comparator cannot register it” (Pitman 1987:340).
In turn, Szechtman and Woody (2004:111) suggest that “the symptoms
of obsessive-compulsive disorder...have what might be termed an epistemic
origin—that is, they stem from an inability to generate the normal ‘feeling
of knowing’ that would otherwise signal task completion.” On the empirical
side, Gentsch et al. (2012:656) found decreased sensory attenuation in OCD,
which was suggested to “explain the tendency of individuals with OCD to con-
tinuously register error signals, and to experience dissatisfaction in outcome
processing.”
The somewhat contrived formulation of OCD, in terms of aberrant salience,
focused on an account of intrusive experience that manifests in overt motor
behavior. Does this explanation hold for intrusive thoughts, images, and expe-
riences in PTSD? A plausible account could proceed along the following lines:
Imagine that, at the point a traumatic event is experienced, there is some par-
ticular configuration of (interoceptive or exteroceptive) sensory inputs in play.
The traumatic event can then induce a one-shot learning of the concomitant
gating policy. When this pattern of sensations is encountered subsequently, it
is extremely difficult to ignore, because sensory information is afforded great
precision. These sensory cues will induce belief updating and the selection of
2
This account is from the PhD thesis by Itzchak (Isaac) Fradkin: “Deficits in processing of
prediction errors in obsessive compulsive disorder: Effects on action, thoughts, learning and
agency,” Hebrew University of Jerusalem, June 2019.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
258 A. C. Roberts et al.
the traumatic narrative or policy that entails overt or covert action. In the latter
setting, action is neither motoric (i.e., mediated by striated muscles) nor auto-
nomic (mediated by smooth muscles) but attentional in nature. In other words,
the gating policy is called up in an obligatory fashion, sometimes described in
terms of modulating sensory and prior precision (Skewes et al. 2014; Ainley et
al. 2016; Powers et al. 2017; Rae et al. 2019a).
This traumatic active inference or learning will induce a recapitulation of
the internal policy or narrative that may enable posterior expectations all the
way down to the sensory levels of perceptual hierarchies. In other words, a
triggering event will breach attentional thresholds and induce a cascade of
hierarchical and sequential processing that recapitulates the sequential narra-
tive associated with the original trauma. The mechanisms behind such fictive
(intrusive) experience are part and parcel of self-evidencing under a generative
model. Common examples here include dreaming, imagination, and the gen-
erative or constructive perceptual processing associated with structure learning
and eureka moments (Hinton et al. 1995; Botvinick et al. 2009; Gershman and
Niv 2010; Tervo et al. 2016; Friston et al. 2017; Gershman 2017).
Based on this account, the intrusive experience induces a gating policy
that prescribes covert (mental) actions that are manifest as internal scene con-
struction and accompanying narratives (Peters et al. 2017; Wilkinson et al.
2017), as opposed to the mostly overt actions considered in the OCD example
above. Clearly, the foregoing account does not offer a qualitative distinction
between intrusive experiences that reflect an adaptive response to trauma and
the psychopathology that results when intrusions are experienced (or manifest)
as maladaptive and persistent. However, the computational account narrows
down the field, in terms of where aberrant inference and learning may be oper-
ating in conditions like OCD and PTSD. Next, we consider the failure of sen-
sory attenuation and subsequent failure to relearn the right sort of attentional
response as a plausible candidate.
Summary
The two working examples of OCD and PTSD were introduced here to make
a key point: the intrusive experience of OCD rests upon aberrant salience
that is secondary to a failure of sensory attenuation; namely, an aberrant top-
down modulation of sensory mappings. In contrast, the PTSD example appeals
only to aberrant precision via a breach of sensory attenuation due to traumatic
learning of a particular attentional set or gating policy. In other words, people
with PTSD may lose the capacity to ignore the irrelevant and be plagued by
breaches of attentional filtering or gating endowed by sensory attenuation. If
one subscribes to these accounts, the conclusion is that the primary pathophys-
iology behind both kinds of intrusive experience is a failure of sensory attenu-
ation that most likely involves interoceptive signals. Interestingly, a failure of
sensory attenuation emerges in computational treatments of other psychiatric
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems Approach to Intrusive Experiences 259
conditions (Skewes et al. 2014; Ainley et al. 2016; Powers et al. 2017; Rae et
al. 2019a); in particular, schizophrenia and autism. [For a review of aberrant
precision and sensory attenuation in psychiatry, see Stephan et al. (2016) and
Friston (2017) for details and references.]
On this account, a minimal but sufficient explanation for intrusive experi-
ence is a failure of inhibitory control inherent in the sensory attenuation. The
key thing that the active inference framework brings to the table is that this in-
hibitory control is not about the contents of perceptual experience, but the pre-
cision or attention afforded this content. From a physiological perspective, this
is important because a failure of inhibition (i.e., a failure of sensory attenuation
or attenuation of sensory precision) may be mediated not by hyper- (or de-)
polarizing neuronal populations but by modulating their excitability or gain.
In turn, this suggests the mechanisms that underwrite the pathophysiology of
intrusive experiences are located either in classical modulatory neurotransmit-
ter systems or the downstream effects on cortical excitability (as mediated by
fast-spiking inhibitory interneuron coupling with pyramidal cells).
In summary, the emerging picture is of a deficit in the neuromodulatory
mechanisms (and dynamics) that implement the top-down control of atten-
tion; namely, its sensory attenuation. A natural corollary is that there may be
as many different forms of intrusive pathologies as there are neuromodulation
mechanisms and projections. Irrespective of this diversity, and the accompa-
nying regional specificity of evidence accumulation schemes in the brain, one
underlying mechanism becomes apparent: the breach of sensory attenuation
(i.e., attentional filtering) by exogenously or endogenously generated cues that
underwrite belief updating about states of the world and our active engagement
with that world. Clearly, in many instances, this intrusion is part of normal per-
ceptual synthesis and subsequent planning. For example, a loud noise is salient
because it offers a person the opportunity to “look over there” and resolve any
uncertainty associated with the surprising sensory signal.
The pathology implicit in the examples above rests on aberrant salience
that maintains irreducible uncertainty incurred through a failure to attenuate
interoceptive signals (as in the case of overt compulsive behavior in OCD).
It can also rest on the failure of sensory attenuation to be attributed to, and
subsequent failure to relearn, the right kind of attentional response to triggers
(as in the case of PTSD). As discussed above, the notion of a breach in sensory
attenuation is a key aspect of higher-order models of intrusive experiences that
consider the evaluation (i.e., the appraisal) of inferred states and subsequent
metacognitive influences. At present, three conclusions follow from the formal
analysis of this section that are remarkably consistent with the treatments of-
fered in other chapters in this volume:
• Intrusive experiences are inherently interruptive in the sense that they
induce a quantitative change in the selection of narratives or sequential
policies, which underwrite overt or covert (mental) action.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
260 A. C. Roberts et al.
Now let us consider the above account from a systems neuroscience perspec-
tive. In this setting, salience can be thought of as an attribute of a cue (i.e.,
internal or external stimulus) deemed important to the individual in a given
context (Uddin 2014; Kahnt and Tobler 2017; Miyata 2019)—it is salient be-
cause of the potential for information gain and thus belief updating. Salience is
distinct from value in that the latter is a valenced or signed currency that varies
monotonically from negative to positive, whereas the former is an unsigned
currency (i.e., something is salient or not). This means that value and salience
are dissociable in terms of what they mean for behavior: both negative and
positive outcomes can be salient in the sense that experiences can change our
beliefs, even if they are unpleasant (Kahnt and Tobler 2017). As such, intrusive
experiences can be thought of as arising from an aberrant processing of inter-
nal and external stimuli with respect to the current (belief) state of the indi-
vidual. This salience misattribution leads to an overemphasis of one thought or
action over the current, ongoing cognitive process and subsequently influences
attentional capture, motivation, and goal-directed cognition. Importantly, the
unsigned nature of salience calculations necessitates that both appetitive and
aversive stimuli can sway the calculations of salience that ultimately influence
behavior.
There are many potential points at which biases can enter salience calcula-
tion. Ascribing salience to a given stimulus at a given time scale (Kennerley
et al. 2011) and within a given context (Heilbronner and Hayden 2016) re-
sults from integration across a wide range of processes, including attentional
(Menon and Uddin 2010), reward (Olney et al. 2018), affective (Etkin et al.
2011), and homeostatic regulation (Craig 2009).
Neurobiological instantiation of both ongoing and intrusive, highly salient
events occurs at many levels of the neuraxis. One highly interconnected hub that
seems to play a major role as an integrator or transmitter of the interoceptive and
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems Approach to Intrusive Experiences 261
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
262 A. C. Roberts et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems Approach to Intrusive Experiences 263
What Is Interoception?
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
264 A. C. Roberts et al.
There are at least three ways in which interoception can impact upon intrusive
experiences:
1. It can provide context which can (a) have an impact on the permissive
threshold for the occurrence of intrusions (discussed above) and (b)
influence or constrain the content of what intrudes.
2. It can affect appraisal and control processes engaged by the intrusive
experience.
3. It can also act as content itself.
Moreover, these can interact to produce a self-sustained cycle of intrusive ex-
periences. In conceptualizing the impact of interoception on intrusive expe-
riences, it is helpful to conceptualize it within a hierarchical or dimensional
framework (see Table 13.1). Lowest in the hierarchy are the levels of physi-
ological arousal (indexed by heart rate, blood pressure, or electrodermal activ-
ity) and the bodily changes governed by homeostatic reflex arcs. These signal
the integrity and arousal state of the body through visceral afferent pathways.
Fluctuations in central signaling of bodily physiology (including both engage-
ment of ascending neuromodulatory systems and representation within pri-
mary “viscerosensory” insula, a cortical level) can thus provide the context
(Pt. 1 from the above list).
As a context, psychophysiological states (e.g., sickness, arousal, and alert-
ness) gate what enters the sensorium (Pt. 1a). For example, a heightened
state of cardiovascular arousal enhances the detection and appraisal of threat
(Garfinkel et al. 2014; Pezzulo et al. 2018) associated with symptoms of anxi-
ety; increased sympathetic electrodermal tone enhances occurrence of tics
in Tourette syndrome (Nagai et al. 2009). In addition, however, a particular
homeostatic context, such as hunger, can motivate relevant intrusions about
food (Pt. 1b; a specific example is given in the next section). Affective state
represents a more elaborated interoceptive context that can again change the
permissible threshold of intrusion.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Table 13.1 Dimensions of interoceptive measurement, adapted after Garfinkel et al. (2015). Psychological dimensions of interoception are given
in boldface; self-referential dimensions are capitalized.
Dimensional level Nature Index measures
EXECUTIVE Behavioral Shifting from interoceptive to exteroceptive attention (e.g., within dual
tasks or between tasks)
METACOGNITIVE Correspondence between subjective self- Receiver operating characteristic curves between task performance and
report and objective performance accuracy rated confidence
Correlational measures of task and confidence scores
Trait measures (e.g., correspondence between task performance and body
perception questionnaire score)
SENSIBILITY Subjective self-report Confidence measures on interoceptive tasks
Questionnaires probing interoceptive sensitivity
Accuracy Objective behavioral performance score Heartbeat detection tasks
Respiratory resistance load detection
Water load task
Balloon dilation of stomach/colon
Preconscious impact Behavioral, neural Cardiac modulation of eyeblink startle
on other processes Cardiac modulation of fear
Respiratory modulation of memory
Afferent signal Neural Visceral afferent nerve recording
Intracranial recording
Heartbeat evoked potential
Respiratory evoked potential
Neuroimaging
Bodily response Organ-level response Heart rate, heart rate variability, tachygastria, blood pressure, glucose, O2
and CO2 levels, etc.
Autonomic psychophysiology
265
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
266 A. C. Roberts et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems Approach to Intrusive Experiences 267
Patients with anorexia nervosa report thoughts, bodily experiences, and mental
images that they consider as involuntary and intrusive to other goals, even
though these may not always be unpleasant in themselves and may, in fact,
constitute most people’s everyday experiences. For example, a patient de-
scribed the feeling of a full stomach as intruding on her mental concentration
(Skårderud 2007:127):
Some days ago, I should have had a meeting with my boss. I was anxious about
this. Then I decided to vomit. I couldn’t stand having the lunch in my stomach. I
cannot have anything in my stomach, because then I cannot concentrate. I need
to be empty to feel alert.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
268 A. C. Roberts et al.
Eating restriction
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems Approach to Intrusive Experiences 269
Balancing stability and flexibility in the brain is critical for individuals to max-
imize exploitation and exploration of their environment. Working memory and
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
270 A. C. Roberts et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems Approach to Intrusive Experiences 271
Though different mechanisms for working memory gating have been pro-
posed (e.g., Wang et al. 2004; Zhu et al. 2018), one influential model has
focused on frontostriatal interactions (Frank et al. 2001). This circuit is sche-
matized in Figure 13.7 (O’Reilly 2006). The corticostriatal model of working
memory gating proposes that the prefrontal cortex supports information main-
tenance, whereas the striatum-pallidal-thalamic pathway implements gating by
regulating what information is allowed in and out of working memory.
Based on this model, spontaneous, unwanted events could be experienced
as intrusions when the gate to working memory is breached and the intru-
sion supplants ongoing working memory processes. Once this occurs, intru-
sive events serve as signals to drive other cognitive processes and actions.
Thus, the integrity of the working memory gating is paramount for mitigating
against intrusive experiences. For example, by preventing an unwanted expe-
rience from updating to working memory or by inhibiting their influence on
output control signals, one could stop the negative cycle of behaviors that can
result from intrusive experiences. These gating mechanisms could be global
(like the fast, inhibitory mechanisms supported by the hyperdirect pathway
that can affect multiple processes simultaneously) or selective, supported by
both the direct and indirect pathways, schematized in Figure 13.7 as the Go
and No-Go pathways. Coordination among multiple corticostriatal loops can
also be a mechanism for working memory operations in separate prefrontal
areas to carry out complex, sequential, and hierarchically structured tasks (for
a review, see Badre and Nee 2018).
Striatum Striatum
Thalamus Thalamus
Go No-Go VA, VL, MD Go No-Go VA, VL, MD
SNr SNr
Excitatory Inhibitory
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
272 A. C. Roberts et al.
The type of gating one selects can be thought of as a gating policy, in the
same sense as defined above. As such, the match of the right gating policy
to the particular dynamics of the situation is a key determinant of success-
ful control (Bhandari and Badre 2018). For example, when confronted with
an unwanted memory, one could deploy a global suppression to prevent it
from entering working memory or instead attempt to selectively input another
thought into working memory in its place. The consequences of these policies
on memory or the ongoing impacts of the triggering event (both in this instance
or in the future) might differ depending on the gating strategy that is selected.
Thus, pathologies could arise as a result of any of the following:
• Items seeking to enter working memory are sufficiently salient or val-
ued and will therefore breach the gating mechanism to update working
memory (breach intrusion).
• Gating itself is weak and thus items access working memory, even if
they are not adaptive or helpful to the individual (permissive intrusion).
• Mechanisms involved in maintaining stability (other than gating) are
too strong and thus do not allow working memory to be updated once
an intrusive experience has occurred.
• The wrong gating policy is selected given the nature or dynamics of
the intrusion.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems Approach to Intrusive Experiences 273
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
274 A. C. Roberts et al.
Summary
Appendix 13.1
Active Inference
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems Approach to Intrusive Experiences 275
For technical readers, expected free energy can be decomposed into an epis-
temic, information-seeking, uncertainty-reducing part (intrinsic value) and a
pragmatic, goal-seeking, instrumental part (extrinsic value). Formally, the ex-
pected free energy for a particular policy π at time τ in the future can be ex-
pressed as in terms of beliefs Q s , o about future states sτ and outcomes oτ:
G , E ln Q s | o , ln Q s | E ln P (o ) . (13.A1)
intrinsic value extrinsic value
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
G
276
Generative model
P o1:W , s1:W , S P ( s1 ) P (S )W P(oW | sW ) P( sW | sW 1 , S )
S
1 P oW | sW Cat ( A)
2 P sW 1 | sW , S Cat (BS ,W )
3 P s1 Cat (D) Factors D s1 BS ,1 s2 BS ,2 s3
4 P S V (G ) (likelihood and empirical priors)
A A A
Q sW | S Cat (sS ,W )
Q S Cat ( ʌ) Approximate posterior o1 o2 o3
ʌ
1
2 3 4
sS ,W V (ln BS ,W 1sS ,W 1 ln BS ,W sS ,W 1 ln A oW ) G
4
1 ʌ V (G ) 3 5 2 5 2 5
sS ,1 sS ,1 sS ,2 sS ,2 sS ,3 sS ,3
D 2 = 3 BS ,1 2 = 3 BS ,2 2 =
5 GS , ,
¦W oS W (ln oS W CW H sS W ) , sS ,1 sS ,2 sS ,3
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Figure 13.A1 A generative model for discrete states and outcomes. Upper left panel: these equations specify a generative model. A generative
model is the joint probability, P, of outcomes or consequences and their (latent or hidden) causes, see first equation. Usually, the model is expressed
in terms of a likelihood (the probability of consequences given causes) and priors over causes. When a prior depends upon a random variable it is
called an empirical prior. Here, the likelihood is specified by a matrix A, whose elements are the probability of an outcome under every combina-
tion of hidden states. Cat denotes a categorical probability distribution. The empirical priors pertain to probabilistic transitions (in the B matrix)
among hidden states that can depend upon actions, which are determined by policies (sequences of actions encoded by π). The key aspect of this
generative model is that policies are more probable a priori if they minimize the (time integral of) expected free energy G, which depends on
prior preferences about outcomes or costs encoded in C and the uncertainty or ambiguity about outcomes under each state, encoded by H. Finally,
the vector D specifies the initial state. This completes the specification of the model in terms of model parameters that constitute A, B, C, and D.
Bayesian model inversion refers to the inverse mapping from consequences to causes (i.e., estimating the hidden states and other variables that
cause outcomes). In approximate Bayesian inference, one specifies the form of an approximate posterior distribution Q. This particular form in
this figure uses a mean field approximation, in which posterior beliefs are approximated by the product of marginal distributions over time points.
Subscripts index time (or policy), italic variables represent hidden states, and bold variables indicate expectations about those states. Upper right
panel: this Bayesian network or graphical model represents the conditional dependencies among hidden states and how they cause outcomes. Open
circles are random variables (hidden states and policies), filled circles denote observable outcomes, and squares indicate fixed or known variables,
such as the model parameters. Lower left panel: these equalities are the belief updates mediating approximate Bayesian inference and action selec-
tion. Lower right panel: this is an equivalent representation of the Bayesian network in terms of a Forney or normal style factor graph. Here the
nodes (square boxes) correspond to factors and the edges are associated with unknown variables. Filled squares denote observable outcomes. The
edges are labeled in terms of the sufficient statistics of their marginal posteriors (see approximate posterior). Factors have been labeled intuitively
in terms of the parameters encoding the associated probability distributions (on the upper left). The circled numbers correspond to the messages
that are passed from nodes to edges (the labels are placed on the edge that carries the message from each node). These correspond to the messages
implicit in the belief updates (lower left). This figure is based on Friston et al. (2017).
277
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
G
278
u1 u2
D s1 Bu1 s2 B u2 s3
A A A
sS ,W V (ln BS ,W 1sS ,W 1 ln BS ,W sS ,W 1 ln A oW )
ʌ V (G )
o1 o2 o3
GS , ,
¦W oS W (ln oS W CW ) H sS W ,
sS ,1 sS ,2 sS ,3
uW max u ʌ [U S ,W u]
Action selection BS ,2
D BS ,1
= = =
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Figure 13.A2 The generative process and model. This figure reproduces the Bayesian network and Forney factor graph of Figure 13.A1; how-
ever, here the Bayesian network describes the process that generates data, as opposed to the generative model of data. This means that we can link
the two graphs to show how the policy half-edge of Figure 13.A1 couples back to the generative process (by generating an action that determines
state transitions). The selected action corresponds to the most probable action under posterior beliefs about action sequences or policies. Here,
the message labels have been replaced with little arrows to emphasize the circular causality implicit in active inference: the real world (red box)
generates a sequence of outcomes that induce message passing and belief propagation to inform (approximate) posterior beliefs about policies (that
also depend upon prior preferences and epistemic value). These policies then determine action, which generates new outcomes as time progresses,
thereby closing the action perception cycle. This figure is based on Friston et al. (2017).
279
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
280 A. C. Roberts et al.
(Srinivasan et al. 1982; Rao and Ballard 1999; Huk and Shadlen 2005; Beck et
al. 2008; Bastos et al. 2012; Egner and Summerfield 2013; de Lafuente et al.
2015; Kira et al. 2015; Shipp 2016). In terms of the parameters of the genera-
tive model, associative plasticity is the corresponding belief update for neuro-
nal connections (Friston et al. 2016).
Of particular interest here are the parameters that link states to outcomes and
states at one point in time to states at the next point in time. In Figure 13.A1,
these are simply matrices of probabilities encoding the likelihood mapping
from states to outcomes A and the transition probabilities to one state to the
next B (which depend upon a particular policy).
Neurobiologically, these matrices play the role of connectivity matrices,
which play an important role in sensory data assimilation and subsequent
planning based on beliefs about the consequences of any action. Furthermore,
each column of these matrices has a precision. Precision, in this instance, re-
flects the fidelity or confidence about the outcome (or subsequent state) given
the current state of the world. A very precise mapping means that we can be
almost 100% confident that this will happen given that state, while a very
imprecise mapping means that all outcomes (all subsequent states) are equally
likely. For discrete space models, one can express the likelihood and priors in
terms of inverse temperature or softmax parameters with the following form,
where ( ) is a softmax function or normalized exponential:
P o |s o ln A
P s | s 1, s ln B (13.A2)
P G .
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems Approach to Intrusive Experiences 281
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
282 A. C. Roberts et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Systems Approach to Intrusive Experiences 283
(a) Neuron
Neuron
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
Time (s) Time (s) Time (s) Time (s) Time (s)
Firing rate
Firing rate
Firing rate
10 10 20
10
0 0 0
1000 1000 1000
60 70 60 70 40 50 60
70
Neu500 30 40 50 Neu500 30 40 50 Neu500 0 0 10 20 30 rial
ron 0 0 10 20 Trial ron 0 0 10 20 Trial ron T
(c)
Switch
Perseveration
Neuron
Memory loss
1
0.8
Proportion of traits
Neuron
0.6
0.4
Neuron
0.2
0
0 1 2 3 4 5 –30 –25 –20 –15 –10 –5 0
Time (s) &KDQJHUHODWLYHWRSK\VLRORJLFDO+7
Figure 13.A4 A computational model of the role of serotonin (5-HT) in the orbito-
frontal cortex in OCD, adapted after Maia and Cano-Colino (2015). (a) Illustration of
the process of entrenchment of patterns of neuronal activity, taken to correspond to
obsessions or, in less pathological cases, “habits of thought.” Each plot represents a
population of neurons; the dots along each line represent the action potentials for one
neuron. The network stochastically develops patterns of activity (“bumps”). Each bump
elicits strengthening of the synapses between the neurons that were active in that bump
through Hebbian learning, thereby developing attractors (orange bands). The more fre-
quently a bump occurs, the more likely it is that it will reoccur (see last three plots).
(b) Effects of reducing serotonin on the tendency to develop and fall into excessively
strong attractors. Under normal circumstances, the network develops bumps at varying
places over time (left panel). Under low levels of serotonin, however, the network tends
to develop excessively strong attractors into which it repeatedly falls (middle panel).
Moreover, there is a dose-response effect, such that reducing serotonin further causes
even stronger attractors to develop (right panel). Increasing glutamate has the same
effect as decreasing serotonin (not shown). (c) Low levels of serotonin cause the at-
tractors to become excessively stable. Simulated activation of a set of neurons elicited
a bump, followed by activation of a different set of neurons. Under normal conditions,
the network’s pattern of activity flexibly shifts to the state represented by the new bump
(blue). Under low levels of serotonin, the network fails to shift to the new bump, result-
ing in perseverative activation of the prior bump (brown). (continued on next page)
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
284 A. C. Roberts et al.
Figure 13.A4 (continued) Importantly, low levels of serotonin increase such perse-
verative errors (brown) without affecting a different type of error in which the network
simply loses the memory of what it was initially representing (green). The latter error,
which is more reminiscent of disorders in which there is difficulty in keeping items in
working memory (e.g., ADHD), is not affected by the serotonin manipulations.
Acknowledgments
Figures 13.A1 and 13.A2, from Friston et al. (2017), are used under the terms of the
Creative Commons Attribution 4.0 International License (CC-BY).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
14
Psychological Interventions
as They Relate to
Intrusive Thinking
Intrusive, Emotional Mental Imagery
after Traumatic and Negative Events
Abstract
Common across psychological disorders, intrusive, emotional mental images are
sensory-perceptual representations that intrude involuntarily into the mind. Mental
health treatments typically focus on entire disorders with multiple symptoms. This
chapter suggests focusing on core clinical symptoms (i.e., intrusive imagery). Existing
psychological therapy techniques (e.g., imagery rescripting) are promising, but under-
lying treatment mechanisms need to be better understood.
Precise treatments and preventions are required. Using the example of psychologi-
cal trauma, this chapter argues that psychological interventions can be developed in the
laboratory: effective experimental analogues of trauma can generate intrusions so that
putative interventions that modulate intrusions can be explored at various mechanis-
tic levels (e.g., molecular, cognitive, social). Examples of targeting “new” (i.e., Day 1
of the traumatic event) memories include a simple cognitive interference intervention
that holds promise for preventing intrusive images after trauma (a behavioral protocol
including Tetris game play). This intervention specifically targets intrusive involuntary
memories while leaving voluntary memory intact. Work on targeting “old” (as of Day 2)
memories is at an earlier stage. Research on reconsolidation update mechanisms appears
valuable in reducing older trauma memories via interference interventions, again with a
behavioral task interference technique. To understand mechanisms across different lev-
els (e.g., molecular, cognitive, or social), mathematical models can aid the identification
of causal mechanisms involved in memory formation. Questions are posed to instigate
discussion of future science-driven psychological interventions for intrusive images.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
288 E. A. Holmes et al.
Introduction
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Current Methods of Psychological Interventions 289
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
290 E. A. Holmes et al.
Care Excellence 2018), as the evidence is less strong for clinical effectiveness.
Drug treatments after trauma are not recommended as a preventive strategy; in
particular, benzodiazepines can worsen symptoms.
Since the inception of CBT in the 1960s, great advances have been made.
For instance, we now have highly effective evidence-based CBT treatments
for full-blown PTSD. This is one of the areas in which we have the best treat-
ments: some of the best PTSD trials have recovery rates of 75%, whereas the
norm in CBT is 50% (Holmes et al. 2014). However, there are still many ways
in which we need to improve psychological treatments (Holmes et al. 2018).
How can we adapt or simplify them to reach more people? How do we under-
stand the critical ingredients in a psychological treatment in so doing? Can we
also “prevent” rather than “cure”? That is, can we prevent intrusive memories
after trauma rather than only have treatments once the full-blown disorder has
been established?
When we think, we can think in the form of words (verbal thought) or men-
tal images (sensory representations in any modality such as visual, olfactory,
or auditory images). The dominant focus in psychological treatment research
and therapy, such as CBT, has been on verbal thoughts. A focus on mental im-
agery is one alternative and may open up opportunities for both research and
treatment innovation.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Current Methods of Psychological Interventions 291
Fear Conditioning
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
292 E. A. Holmes et al.
neutral cues with threat can become maladaptive when such conditioned fear
fails to extinguish long after the danger has passed or overgeneralizes to a
wider range of contexts.
Fear-conditioning paradigms allow for an investigation of the emergence,
persistence, and resurgence of maladaptive fear responses (Pittig et al. 2018). It
has been suggested that intrusion, hyperarousal, and hypervigilance symptoms,
which characterize PTSD, may arise as a result of conditioned fear responses
(for a review, see Norrholm and Jovanovic 2018). Note, however, that fear-
conditioning experiments have mainly investigated arousal or hypervigilance
(e.g., skin conductance, startle reflex–fear responses) that are mostly relevant
to anxiety disorders.
Intrusions, however, are the hallmark feature of PTSD, which is no longer
classified as an anxiety disorder but as a trauma- and stressor-related disor-
der in DSM-5 (American Psychiatric Association 2013). As fear-conditioning
paradigms do not specifically account for the image-based episodic nature of
intrusive thoughts (Visser et al. 2018), it remains an interesting open question
whether this paradigm could also be used to generate intrusive memories of
a stressor in the laboratory. A more ecologically valid and clinically relevant
experimental model may be needed to generate and study intrusions per se in
the laboratory.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Current Methods of Psychological Interventions 293
over the course of daily life (James et al. 2016a). This method records intrusion
frequency data over longer time frames and carefully matches intrusions to the
trauma film (participants usually record whether they had an intrusion or not
for several time periods per day over the course of one week and briefly de-
scribe each intrusion’s content). Mixture models can be a useful tool to analyze
such diary data because they model intrusion and non-intrusion data differently
(see discussion on mechanisms and mathematics below for further details).
Additional Methods
Watching visual stills of distressing content (e.g., injured people) has also been
shown to generate intrusions two days later (Battaglini et al. 2016). In addition,
listening to negative arousing stories while watching a slide show of pictures
can generate negative emotional memories (Galarza Vallejo et al. 2019).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
294 E. A. Holmes et al.
There has been an increased interest in the impact of social factors on emo-
tion regulation. Both human and nonhuman experiments have shown that the
presence of another during an aversive experience may work as a buffer by
reducing fear responses (Thorsteinsson and James 1999; Mikami et al. 2016).
Experiences of social support could increase the process of learning what is
safe in the environment (social safety learning) through social support interac-
tions, which in turn decrease stress reactivity to stressful experiences (Ditzen
and Heinrichs 2014).
After a psychologically traumatic event, social support (i.e., supportive
interactions with family and friends) is believed to be associated with hav-
ing fewer posttraumatic cognitions (e.g., trauma-related thoughts and beliefs),
which in turn is associated with PTSD symptoms (Woodward et al. 2015).
These results signal a need to investigate social interactions and social support
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Current Methods of Psychological Interventions 295
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
296 E. A. Holmes et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Current Methods of Psychological Interventions 297
Figure 14.1 Diagram depicting how different memory systems may represent various
aspects of a traumatic event (e.g., sexual abuse by a piano teacher). In general, clinically
beneficial interventions should aim to target the maladaptive involuntary expression of
trauma memories (e.g., intrusive memories) while preserving its voluntary recall (e.g.,
ability to testify in court). Adapted after Visser et al. (2018).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
298 E. A. Holmes et al.
Table 14.1 Overview of different aspects of trauma memory that can be targeted and
associated research approaches, from animal models (bottom) to clinical populations
(top). Left: different levels at which trauma can be modeled. Middle: potential targets
for intervention. Right: memory readouts: (1) occurrence of intrusive images (e.g., diary,
provocation task), (2) event details (e.g., interview), (3) learning episode details (e.g.,
recognition test), (4) self-report of symptoms, (5) rating of subjective distress, (6) un-
conditioned stimulus expectancy, (7) attentional bias, (8) approach/avoidance behavior,
(9) noninvasive physiology, (10) invasive physiology. Note: voluntary memory recall
(e.g., trauma details) can be measured in humans but is not the key clinical target of a
treatment. Adapted after Visser et al. (2018).
Memory
Level Targets
readouts
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Current Methods of Psychological Interventions 299
can become malleable again (e.g., Sara 2000; Alberini 2005; Nader and Hardt
2009). The restabilization of a memory is a putative process termed memory
reconsolidation (Nader et al. 2000). This process is dependent on de novo pro-
tein synthesis; interventions that directly or indirectly target this process thus
have the potential to change maladaptive emotional memories (Milton and
Everitt 2012), including those giving rise to intrusive images of trauma. Figure
14.2 depicts different time windows of memory malleability.
Different interventions can interfere with the reconsolidation of a memory
on different levels. On a molecular level, fear-conditioning studies in rodents
have shown the potential of pharmacologically disrupting one-day-old (Nader
et al. 2000; Lee et al. 2006; Ortiz et al. 2015) and even one-month-old (Gräff
et al. 2014) memories, resulting in a persistent attenuation of conditioned fear
responses. Even though less consistent (Lonergan et al. 2013), these types of
findings have been translated to studies in humans, where the beta-adrenergic
antagonist propranolol was used to disrupt pharmacologically one-day-old fear
memories (Kindt et al. 2009) as well as much older memories; that is, fear
memories that underlie simple phobias for spiders (Soeter and Kindt 2015). On
a cognitive level, behavioral interventions, including extinction after memory
retrieval procedure, have been shown to attenuate one-day-old fear memory in
rodents (Monfils et al. 2009). This finding has been translated to one-day-old
(Schiller et al. 2010) and one-week-old (Steinfurth et al. 2014) memories in
humans, and more recently also to older memories such as those underlying
simple phobia for spiders (Björkstrand et al. 2016) and snakes (Telch et al.
2017). For further details, we refer the reader to recent overviews on mem-
ory reconsolidation literature (Lee et al. 2017; Elsey et al. 2018; Monfils and
Holmes 2018).
With regard to intrusive memories, a visuospatial interference intervention
administered after a reminder cue was effective in reducing intrusive memories
for established (24-hour-old) memory of experimental trauma (James et al.
2015). In this study, individuals who underwent a memory reactivation pro-
cedure and performed an intervention, including Tetris game play, had fewer
intrusive memories than a no-reactivation/no-Tetris group. More recently, two
studies used a similar reactivation and cognitive task interference procedure,
administered three days (Kessler et al. 2020) or four days (Hagenaars et al.
2017) after trauma film viewing; again, a reduction in subsequent intrusive
memories was demonstrated. While both studies showed that an active control
condition (verbal task) also reduced intrusions compared to a no-task control,
in one study the effect was significantly larger for the visuospatial interfer-
ence intervention compared to a verbal control task (Kessler et al. 2020).
Interestingly, and again in line with separate trace theories (Lau-Zhu et al.
2019), both Kessler et al. (2020) and James et al. (2015) showed that the inter-
vention left voluntary memory (i.e., performance on a recognition task) intact.
Still, more work is warranted.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
300
Retrieval Retrieval
Memory susceptibility
Time
Unsuccessful Successful Successful intervention (blockade soon Unsuccessful Successful
intervention intervention after encoding, within the encoding intervention (blockade intervention
(pre-trauma) (blockade during context, or preceded by cue that only, no retrieval cue, (blockade preceded
encoding) directs internal attention to trauma) or vice versa) by retrieval cue)
Figure 14.2 In the hours after an experience, memories are believed to go through an initial labile phase before being stored into stable long-term
memory (i.e., consolidation). The purple arrow depicts different time intervals with respect to the encoding of an aversive episode. The gradients
below indicate the putative processes of memory encoding and consolidation that occur during these different intervals. Recent insights suggest
that certain aspects of memories, including the intrusiveness, are not necessarily permanent. Instead, they may become transiently malleable upon
reactivation, rendering them susceptible to interference or updating before returning to a fixed state, a process referred to as “memory reconsolida-
tion.” This offers a second window of opportunity to interfere with consolidated memories (shown as yellow background shades), making them
less intrusive. Successful interventions (blue arrows) need to be timed such that the blockade interferes with memory when it is in an active, sus-
ceptible state (indicated by the dotted yellow line), either in the first hours after trauma or at later time intervals after a retrieval procedure (e.g.,
reactivation through reminder cues). In the first hours after an experience, blockade procedures may also need to be preceded by cues that orient
attentional resources to the event in order for procedures to successfully interfere with it (e.g., when the intervention is delivered in a context other
than the one in which the trauma occurred). Unsuccessful interventions, timed when memories do not yet exist or are in a fixed state (i.e., not
recently retrieved), are shown. Adapted after Visser et al. (2018).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Current Methods of Psychological Interventions 301
Real trauma memories are typically stronger and broader than aversive
memories formed in the laboratory. Finding the optimal conditions and re-
minder cues to reactivate and render a memory labile (a first step for suc-
cessful interference) is assumed to be much more challenging (Monfils and
Holmes 2018) for real memories of trauma. Yet, a recent study on inpatients
with complex trauma (Kessler et al. 2018) has shown promise in attenuat-
ing the intrusiveness of memories for old trauma some years previously.
Twenty patients monitored the occurrence of intrusive trauma memories over
the course of their admission (5–10 weeks). Weekly interventions involved a
memory reminder for a selected (particularly distressing) memory, followed
by 25 minutes of playing Tetris. A within-subjects multiple baseline design
was used, in which the pre-intervention period was varied. Further, some in-
trusions were never targeted by the intervention. The frequency of targeted
intrusive memories reduced, on average, by 64% from baseline to the pos-
tintervention phase, whereas never-targeted intrusions reduced in frequency,
on average, by 11% over a comparable time period. This shows that even
persistent, older memories of real-life trauma can be changed using memory
interference techniques.
Despite its clear promise for clinical translation, it should be noted that a
number of potential limitations and boundary conditions of reconsolidation-
based clinical applications have been raised (Treanor et al. 2017; Monfils and
Holmes 2018). Moreover, at present, it is not possible to attribute conclu-
sively therapeutic gains to reconsolidation mechanisms (Elsey et al. 2018).
Nevertheless, the notion of memory plasticity has proved useful in inspiring
new avenues for intervention for older memories of trauma (Figure 14.2). Of
particular interest to our current discussion is the potential to modify intrusive
features of memory during time windows of memory plasticity.
To be able to interfere with older trauma memories, the memory trace has to
be activated in working memory via a reminder cue. According to reconsolida-
tion theory, there is an optimal duration for a reminder cue. When memory is
retrieved via a brief learning experience (e.g., one unreinforced conditioned
stimulus) it enters a labile state. However, if retrieval is prolonged (e.g., four
unreinforced conditioned stimuli), the memory might enter a “limbo state,”
and if it is prolonged further, finally extinction. In short, if the reminder cue
“dosage” (duration, instances) is too little, nothing happens (no labilization);
if it is too big, the memory may enter a “limbo state” (nothing happens) or
extinction learning—a new inhibitory trace is formed, fear/distress diminishes,
but this effect may be temporary as it does not alter the original emotional
memory trace (Lee et al. 2006; Merlo et al. 2014; Sevenster et al. 2014). All
three possibilities (no labilization, limbo state, extinction learning) are differ-
ent than the reconsolidation state, so the optimization of the retrieval procedure
follows an inverted U-shape.
From a clinical perspective, experiencing an intrusion may even offer an
opportunity to interfere with reconsolidation of this memory by engaging in
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
302 E. A. Holmes et al.
a competing cognitive task (e.g., playing Tetris) within a specific time frame
after the intrusion occurred (minutes). However, the question is whether spon-
taneous retrieval by means of experiencing an intrusion induces the required
reconsolidation state, or instead any of the other states, in which case one can-
not really interfere with it. This is an important empirical question which has
yet to be tackled.
Recently, an experimental paradigm has been developed to capture intru-
sive memories as they occur in the fMRI scanner (Clark et al. 2015). After
viewing scenes of traumatic events (trauma film paradigm; James et al.
2016a), particular scenes then intrude for an individual. A specific intrusive
memory is triggered in the scanner by a reminder cue. The first results of
experiencing an intrusive memory are shown in Figure 14.3. Understanding
the neural mechanisms of experiencing an intrusive memory may yield
insights for treatment (e.g., for neuromodulation strategies that could be
combined with behavioral interference techniques, such as our Tetris proce-
dure). Colleen Hanlon (this volume) discusses transcranial magnetic stimu-
lation (TMS). It is possible that understanding the neural mechanisms of
an intrusive event (e.g., Clark et al. 2015) alongside associated multivoxel
pattern analysis (e.g., Clark et al. 2014b) will inform how best to apply
TMS during an interference procedure (e.g., Kessler et al. 2018) to reduce
the occurrence of intrusive memories. However, to date this has not been
attempted.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Current Methods of Psychological Interventions 303
As discussed above, memory consolidation and memory trace are not fixed
(Müller and Pilzecker 1900; McGaugh 2000). Neural changes and reorganiza-
tion operate through the time period from the perception of an event to the later
memory retrieval. Given that memory trace is probabilistic (changes between
states) and dynamic in time (and across spatial organization in the brain), we
develop mechanistic frameworks to link across scales of organization to cogni-
tion. While the molecular basis of memory involves epinephrine and cortisol
and protein synthesis to affect changes in synaptic consolidation (Dudai 2004),
scaling this up to focus on the neural mechanism of systems-level consolida-
tion requires appropriate tools. We argue that this is best achieved within the
frameworks and architectures of mathematics.
To bridge the gap between (intrusive) memory consolidation and treatment
through a mechacognitive lens, we use a mathematical framework coupled to
data analysis. Here, our framework focuses on intrusive memory consolidation
and cognitive interference interventions including a visuospatial task (Figure
14.4). In this way we consider how perception (bringing to mind) of a trau-
matic event (zM) following an orientation cue (also called reminder cue for
consistency with the reconsolidation literature) might lead to memory con-
solidating into an intrusive memory (iM) or a more neutral task memory (tM)
following a task (T).
To formalize this, we consider a discrete-time Markov chain (Kemeny and
Snell 1960) where memory moves through a series of states, and the probability
of moving from one state to the next is only dependent on present state. Discrete-
time Markov chain models underwrite many common state space models of
data (e.g., latent variable models, hidden Markov models, and Markov decision
processes). In our current (illustrative) application, we assume that the states
are directly observable and focus on the probability transitions from one state
to another and their implications for understanding therapeutic interventions.
Markov chains can be described by a directed graph where edges are the proba-
bilities of moving between states and vertices represent the states (Figure 14.4).
The directed graph illustrates the steps needed to move from trauma memory
state to consolidated task memory state. For instance, this approach allows clear
assumptions to be formulated; one assumption is that trauma memory needs to
be in a labile state before tasks can be undertaken and affect consolidation of the
task memory. We emphasize that this is all a probabilistic process as we learn
how to scale up to aggregate memory processes, driven from the molecular,
short- or long-term scale to the system level (Albo and Gräff 2018) to a level of
organization at the cognitive scale.
As noted, in this exemplar we consider four states: initial trauma memory
state, labile memory, intrusive memory, and consolidated neutral task memory.
This can be represented by the following transition matrix (N):
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
304
(a) 0 to 3 seconds (Intrusive memory involuntary recall > Control button press)
1.7 3
3 to 6 seconds (Intrusive memory involuntary recall > Control button press) 1.7 3
6 to 9 seconds (Control button press > Intrusive memory involuntary recall) 1.7 3
Right Left Right Left Right Left Right Left Right Left Right Left
Z = –16 Z = –4 Z=8 Z = 20 Z = 32 Z = 44
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
(b) Intrusion Control
Middle frontal Superior frontal Operculum
0.6 0.6 0.6
0.4 0.4 0.4
0.2 0.2 0.2
0 0 0
0 3 6 9 3 6 9 –3 3 6 9
–0.2 –3 –0.2 –3 –0.2 0
0
Figure 14.3 Intrusive memory involuntary recall. Top: Whole-brain analysis showing the increased blood oxygen level-dependent (BOLD)
response for intrusive memory involuntary recall versus control button press group at the two time bins (0–3 s and 3–6 s in relation to the button
press): note the significant differences in activation and the one time bin (6–9 s) of increased BOLD response for the control button press group
versus intrusive memory involuntary recall. Bottom: Region-of-interest profile plots of the signal change observed across each time bin from –3
to +12 s in relation to the button press. Intrusive memory involuntary recall signal change activation is shown in red; control button press signal
change activation in blue. Values are means; standard deviations represented by vertical bars. IFG: inferior frontal gyrus. Adapted after Clark et
305
al. (2015).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
306 E. A. Holmes et al.
0 p1 1 p1 0
p3 p2 1 p2
0
N 1 p3 (1 p3 ) 1 p3 , (14.1)
0 0 1 0
0 0 0 1
p3
p1
Trauma state Labile memory
1 – p1 p2 1 – p2
p = 1.0 p = 1.0
Figure 14.4 Markov chain model: Directed graph representing a Markov chain
framework for exploring intrusive and more neutral task memory consolidation. Ar-
rows (edges) represent transition probabilities between states, and boxes (nodes) rep-
resent different memory states. p1 represents the probability that the reminder cue is
successful. We describe this probability as 1/(1 + exp(–α)), where α is the strength of
the reminder cue. p2 represents the probability that task intervention is unsuccessful and
parameterized here as 1/(1 + T), where T is the strength of the task intervention. p3 is the
probability maintained in a labile state. Once memories enter an intrusive memory or
neutral task memory they are fixed in these states (p = 1.0).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Current Methods of Psychological Interventions 307
1 1
Pr iM | trauma 1
1 exp 1 exp 1 T
(14.3)
1 exp T
.
1 exp T exp T
The terms that contribute to this conditional probability can also be determined
intuitively by looking at Figure 14.4 and tracing the paths from the trauma
state to the intrusive memory state, accumulating the probabilities along all
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
308 E. A. Holmes et al.
(a) (b)
1.0 1.0
labile memory states (p3)
Probability of staying in
Probability of staying in
2.0
1.75
0.6 0.6
1.50 1.5
1.25
0.4 0.4 1.0
1.00
0.5
0.75
0.2 0.2
0.50
0.0 0.0
0 1 2 3 4 5 0 1 2 3 4 5
Strength of reminder cue (Į) 6WUHQJWKRIUHPLQGHUFXHĮ
Figure 14.5 Markov chain analysis of transient memory states. (a) Expected times
and (b) variances in expected times in labile memory states in terms of strength of
reminder cue (α) and the probability of staying in the labile memory state (p3). Strong
reminder cues and high probability of maintaining a memory in a labile state favor long
retention times in this transient (labile) memory state. However, most uncertainty in ex-
pected retention times is observed for low reminder cue strengths and high probability
of maintaining a memory in a labile state.
direct and indirect paths. The marginal change in the probability of iMs is more
sensitive to changes in the probability of task success (p2) than the reminder
cue success probability (p1), as the following inequalities for the change in
d(Pr(iM | trauma)/dpi can be shown to hold:
d Pr iM | trauma d Pr iM | trauma
p1 p2 1 . (14.4)
dp2 dp1
The above inequality holds from our parameterization of p1 and p2:
1 1
1 , (14.5)
1 exp 1 T
which, again, are the probabilities along the path from trauma to task memory
consolidation (Figure 14.4). The probability that a task memory consolidates is:
Pr tM | trauma p1 1 p2 , (14.6)
which is the probability that the reminder cue is successful (p2), and the task
intervention is successful (1 – p2). Again, the terms that contribute to this con-
ditional probability can also be determined intuitively from Figure 14.4, trac-
ing paths from the trauma state to the consolidation state, and accumulating the
probabilities along all direct and indirect paths. Two key predictions from this
analysis are as follows:
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Current Methods of Psychological Interventions 309
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
310 E. A. Holmes et al.
p 1 p exp u if j 0
Pr iM j iM
u exp u . (14.7)
1 p if j 0
iM !
Conclusions
Recent findings on intrusive mental images, reviewed in the first part of this
chapter, as well as the mathematical model on intrusive memory consolidation
and visuospatial interference interventions, introduced in the second part, give
reason to take a step back and think about how current psychological interven-
tions might be improved. We have proposed that to progress in this regard, we
need to know more about specific processes involved in intrusive thinking and
adopt a targeted treatment approach (Iyadurai et al. 2019). We conclude by
raising the following questions to invigorate discussion on how we can make
future psychological interventions more precise and effective.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Current Methods of Psychological Interventions 311
0.5
0.20
0.4
0.15
0.3
Density
Density
0.10
0.2
0.05 0.1
0.00 0.0
0 5 10 15 20 0 2 4 6 8 10
Number of intrusive memories Number of intrusive memories
Figure 14.6 Density distribution of intrusive memories (adapted after Iyadurai et al.
2018) for (a) the attention placebo control group and (b) the active cognitive interven-
tion (reminder cue + visuospatial interference task) group. Statistical analysis on (b),
using a zero-inflated Poisson intercept-only model (Equation 14.7), reveals that the
nonzero counts have an intercept that is significantly different from zero (intercept =
0.908 ± 0.144 (SE), z-value = 13.72, p < 0.001), while the zero counts do not (intercept
= 0.0006 ± 0.144, z-value = 0.0004, p = 0.997). The expected probability of no intrusive
memories in the active intervention group, determined from Equation 14.7 and across
this group of patients, is 0.542.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
312 E. A. Holmes et al.
What if we are able to prevent new intrusive memories as well as tackle older
ones? In addition to preventing the consolidation of “new” intrusive memo-
ries with visuospatial interference interventions directly after a traumatic event
(the same day), we argue that a similar type of intervention could be adapted
to target “old” intrusive memories. Importantly, when targeting older trauma
memories (24 hr to several years after the traumatic event), studies have indi-
cated that an approximately 10-minute gap has to be added between reminder
cue and intervention, supposedly to make the memory trace malleable (Agren
et al. 2012; Schiller et al. 2013; James et al. 2015; Kessler et al. 2018, 2020),
although more research on this gap is needed. Methodologically, this opens ex-
perimental designs to study visuospatial interference interventions. For exam-
ple, where there are older intrusive memories of several different events, one
could target single intrusive images one after the other (i.e., one at a time) and
compare frequency and distress of targeted and nontargeted intrusions over
time (Kessler et al. 2018). Targeting old intrusive memories is clearly impor-
tant for vulnerable patient groups (e.g., refugees). Creating rational approaches
that look more like computer game play may be useful for those who do not
seek traditional psychological help because of perceived stigma. Developing
a brief, easily accessible, and nonstigmatized cognitive intervention that could
potentially be self-administered would fill an important gap to reach such vul-
nerable patient groups (Holmes et al. 2018).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Current Methods of Psychological Interventions 313
Acknowledgments
EAH receives grant support from the Oak Foundation, The Lupina Foundation, and the
Swedish Research Council (2017-00957). RMV is supported by a Netherlands Organi-
zation for Scientific Research (NOW Veni) grant (016.195.246). Figures 14.1 and 14.2,
from Visser et al. (2018), are used under the terms of the Creative Commons Attribution
4.0 International License (CC-BY).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
15
Pharmacological
Interventions as They Relate
to Intrusive Thinking
Harriet de Wit and Anya K. Bershad
Abstract
Intrusive thoughts are features of numerous psychiatric disorders. They vary widely in
form, duration, frequency, and severity. They are associated with disorders with widely
differing pathophysiology, and they are likely to respond to different pharmacological
treatments. It is possible that intrusive thoughts represent a cross-diagnostic symptom
that can be a pharmacological target in their own right, separate from the associated dis-
order. This chapter considers the challenges in studying intrusive thoughts as a separate
entity. It examines intrusive thoughts that are symptoms of several different psychiat-
ric disorders and reviews the medications that have been used to treat them. It holds
that relatively little is known about the effects of psychiatric medications on intrusive
thoughts, either within disorders (separate from other symptoms) or across disorders. A
wide range of medications is used to treat intrusive thoughts that target different neu-
rotransmitter systems. In addition to the psychopharmacological armamentarium, new,
single dose treatments (e.g., ketamine, psilocybin, and MDMA) have emerged that may
specifically address intrusive thoughts across the psychiatric spectrum. In conclusion,
possible directions are discussed for identifying subcategories of intrusive thoughts that
could advance research and treatment in this area.
Introduction
While most of the time we have a sense of control over our own thoughts,
sometimes it seems as if our thoughts control us. Under these circumstances,
thoughts appear to assert themselves into our consciousness, influencing mood
and sometimes behavior. These unwanted images and urges, or “intrusive
thoughts,” can interfere with everyday functioning (Clark 2005). In an early
definition, Rachman (1981) identified three criteria for intrusive thoughts: (a)
the thought interrupts ongoing activity, (b) the thought is recognized to be
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
316 H. de Wit and A. K. Bershad
of internal origin, and (c) the thought is difficult to control. When intrusive
thoughts become severe enough to interfere with normal function, they be-
come clinically significant psychiatric symptoms and thus potential targets for
treatment. In this chapter we review several psychiatric disorders in which in-
trusive thoughts are a prominent symptom and summarize the medications that
have been used to treat these disorders. We also comment on future directions
of pharmacological treatments for this symptom.
In the context of psychiatric disorders, intrusive thoughts vary widely in
their nature, form, frequency, and controllability, and this variance can make
them difficult to study as well as to treat. Intrusive thoughts may occur as part
of normal human experience, such as in the case of grief or intense romantic
love, or they may indicate a serious psychiatric disorder such as psychosis
(Table 15.1). They vary in the degree to which they interrupt normal thinking
and activity, whether they are perceived to originate from within the individual
or from outside, and the extent to which the individual feels they can be con-
trolled. Some intrusive thoughts may be experienced as pervasive emotional
experiences, such as worry or rumination in anxiety and depression, whereas
others may be experienced as either spontaneous intrusive experiences, such
as compulsions in obsessive-compulsive disorder (OCD). Others are elicited
by discrete environmental stimuli, such as triggers in posttraumatic stress dis-
order (PTSD) or craving elicited by cues in drug users. Intrusive thoughts also
differ in content along several modalities, including verbal, visual imagery,
and sensorimotor phenomena (e.g., urges and compulsions) and the degree to
which they can be suppressed or controlled by various behavioral procedures.
Although studying these dimensions may shed light on the psychological and
neural processes underlying intrusive thoughts, there have been few attempts
to categorize intrusive thoughts along any of these dimensions.
An obvious follow-up to recognizing this variability is to ask: To what
extent do pharmacological or behavioral treatments depend on the nature of
the intrusive thoughts? Certain drugs, for example, may be effective for intru-
sive thoughts related to depressive rumination and suicidality, whereas others
may more effectively target stimulus-elicited urges. Careful examination of
intrusive thoughts as separate entities may reveal shared mechanisms across
diagnoses, which can be revealed by focusing on medications that are effica-
cious across disorders. Naltrexone, for example, is used for opioid use as well
as for OCD, and aripiprazole is used for suicidal thoughts and for gambling.
Such analysis is consistent with the RDoC approach, designed to identify
the neurobiological and cellular mechanisms underlying psychopathological
states. Increasingly, researchers have focused on the neural features of intru-
sive thoughts. For example, Popa et al. (2016) report that stimulation in the
dorsolateral prefrontal cortex in patients with epilepsy produces the persistent
thoughts that often presage frontal seizures. Other studies (e.g., Hellerstedt
et al. 2016) have used a Think/No-Think procedure to track neural activity
related to unwanted memories using electrophysiological measures. Kühn et
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Pharmacological Interventions 317
Table 15.1 Examples of intrusive events that occur in different psychiatric disorders.
al. (2013) used fMRI to show that there was greater activity in brain regions in-
volved in language production in healthy individuals who reported high, com-
pared to low, habitual tendencies for intrusive thought. This focus on intrusive
thoughts as entities in their own right provides an opportunity in the future to
investigate pharmacological interventions that target such thoughts.
Here we review the main medications that are used for psychiatric disor-
ders for which intrusive thoughts are a major feature (Table 15.2). We separate
the medications into categories based on first-line treatment, second-line treat-
ment, and experimental approaches. The medications listed under the first two
categories are considered standard care and were drawn from the online clini-
cal decision support resource site UpToDate. The medications listed under ex-
perimental approaches were drawn from a review of the literature (PubMed).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Table 15.2 Drugs used to treat psychiatric disorders with intrusive thoughts in first-line treatment, second-line treatment (where several random-
318
ized control trials have shown strong evidence), and experimental approaches (where some small studies may show effects). MAOI (monoamine
oxidase inhibitor), MDMA (3,4-methylenedioxymethamphetamine), SNRI (serotonin and norepinephrine reuptake inhibitors), SSRI (selective
serotonin reuptake inhibitors), TCA (tricyclic antidepressants).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Table 15.2 (continued)
*Psychosis may be part of many different disorders, some of which are treated differently than the above.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
320 H. de Wit and A. K. Bershad
In the sections below, we group the intrusive thoughts into four broad catego-
ries based on their phenomenology, known biological or environmental ori-
gins, and involvement of emotional states:
1. Behavioral: intrusive thoughts that take the form of urges and compul-
sions to engage in repetitive or unwanted actions.
2. Affective: intrusive thoughts that appear to emerge from valenced emo-
tional states, such as anxiety or depression.
3. Substance-induced: intrusive thoughts that are related to psychoactive drugs.
4. Cognitive: intrusive thoughts that involve delusional thoughts and
hallucinations.
Intrusive thoughts are rarely a primary target of treatment, and when they are,
they are usually treated with psychosocial interventions (e.g., Clark 2005;
Ainsworth et al. 2017; van Schie and Anderson 2017; Rebetez et al. 2018;
Iyadurai et al. 2019). However, as we note below, a small handful of studies
have used pharmacological techniques to reduce intrusive thoughts.
Behavioral
For several disorders, intrusive thoughts take the form of strong urges to en-
gage in harmful, apparently unnecessary, or socially unacceptable actions. This
form is a key diagnostic symptom of OCD (Clark and O’Connor 2005) and
may include thoughts about engaging in inappropriate sex acts, violence or
harm to others, or thoughts related to family members, children, or death. They
may also take the form of fear of germs or contamination as well as discomfort
at having things out of order, accompanied by urges to act (e.g., cleaning or
washing, repeated checking or repeated counting). In the case of body dysmor-
phic disorder, patients may be obsessively preoccupied with a particular aspect
of their appearance and make repeated efforts to alleviate this discomfort by
altering their bodies. Patients with OCD may worry about engaging in socially
inappropriate behaviors, such as touching or kissing someone inappropriately,
hurting someone, or engaging in actions that go against the individual’s value
system. OCD patients report escalating anxiety as they resist urges to act and a
temporary sense of relief after acting on the urge. To varying degrees, intrusive
thoughts that relate to urges to engage in socially unacceptable behaviors are
also features of other impulsive control disorders, such as eating disorders,
gambling, pyromania, kleptomania, and intermittent explosive disorder. In
each of these cases, patients become preoccupied by thoughts of engaging in
an inappropriate behavior; these thoughts are difficult to control and can inter-
fere with their normal function.
For each of these disorders, selective serotonin reuptake inhibitors (SSRIs)
are the first- or second-line of treatment. The behavioral processes through
which SSRIs relieve OCD and associated disorders have been studied in some
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Pharmacological Interventions 321
depth, but are still not understood. Drugs may reduce anxiety and dampen
responses to stimuli that trigger the obsessions, or they may modulate the ob-
sessive thoughts themselves. Tricyclic antidepressants are also used for OCD,
and second-line treatments include other serotonin and norepinephrine reup-
take inhibitors (SNRIs) with less serotonergic activity (Hirschtritt et al. 2017).
Cognitive behavioral therapy is an accepted nonpharmacological treatment for
OCD, and there is some support for adjuvant use of neuroleptics, which may
be especially helpful in OCD patients with tics, though may not act to reduce
obsessive thoughts (Bloch et al. 2006).
Affective
Intrusive thoughts are key features of several affective disorders, which are
characterized by strong emotional states such as depression, anxiety, or mania.
Patients with major depressive disorder report ruminative intrusive thoughts
that are congruent with their negative mood states, such as self-deprecating
thoughts of worthlessness or guilt, delusions of guilt, or paranoia in psy-
chotic depression. Patients with generalized anxiety disorder report intrusive
thoughts that are congruent with their anxious mood states, such as worries
about life, including finances, family, work performance, or accomplishing
everyday tasks. In another disorder in which anxiety is a prominent feature,
PTSD, intrusive thoughts relate directly to the experienced trauma, typically
in the form of vivid memories related to the traumatic event, including the
people, context, and emotions experienced. Individuals with PTSD exhibit a
heightened sensitivity to cues related to the trauma, which may generalize to
other stress-related cues. In PTSD, intrusive thoughts are typically associated
with strong negative affect, including both anxiety and depression. The intru-
sive thoughts themselves are highly distressing and may be accompanied by an
urge to act, in the form of fighting or fleeing from the situation. They are also
a feature of intermittent explosive disorder in which patients experience explo-
sive outbursts of anger and violence. These may be spontaneous or elicited by
inconsequential events perceived as provocation. The final category of mood
disorders, those including symptoms of mania, may be accompanied by strong
urges or recurrent thoughts that lead to impulsive behaviors, such as financial
spending or risky sexual behavior, but it is not clear to what extent these urges
are distressing to the individual experiencing them.
In major depressive disorder, intrusive thoughts which occur as part of
ruminations are typically treated with SSRIs. While ruminative thoughts are
not part of the diagnostic criteria for a major depressive episode, they fre-
quently occur as a part of the disorder. Commonly used instruments, like the
Hamilton Depression Rating Scale and the Beck Depression Inventory, do not
directly assess ruminative thinking, so little is known about the direct effect of
antidepressant medications on these types of thoughts. Future research may
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
322 H. de Wit and A. K. Bershad
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Pharmacological Interventions 323
with drugs that act on the endocannabinoid system, but such medications have
not yet reached clinical application.
Drug Induced
In substance use disorders (SUDs), intrusive thoughts are a key feature. They
typically take the form of preoccupation with drug use, reactivity to drug-related
stimuli, and craving or urges to use drugs. Indeed, “craving” was recently added
to DSM-5 (American Psychiatric Association 2013) as a symptom in the diag-
nostic criteria for SUDs, and it is among the most frequent symptom reported by
drug users attempting to quit. Cravings are problematic not only because they
predict or presage the occurrence of actual drug use, but also because they can be
sufficiently distressing to drug users to be a target of treatment in their own right,
independently of actual drug use (Green and Ray 2018). The origin of drug-
related intrusive thoughts in SUDs is not fully understood and may be multidi-
mensional, reflecting both underlying physiological conditions and responses to
Pavlovian conditioned stimuli. For example, craving is a central feature of acute
withdrawal from a drug after regular use, and it is also elicited by drug-related
cues in the environment or memories of drug use. Cravings may be elicited in
abstinent users, even long after withdrawal, by drug-related stimuli, by positive
or negative emotional events, or by ingestion of a small amount of the drug it-
self. In other cases, cravings or urges occur without any definable precipitant.
Whether all these forms of drug-related intrusive thoughts and cravings reflect a
single underlying neural process remains unknown. Notably, however, there are
intrusive thoughts specific to each drug, and withdrawal from one drug does not
induce craving for other drugs.
Although intrusive thoughts are a key feature of SUDs, they are not usually
selected as a separate target symptom for pharmacological treatment. Thus, as
with other psychiatric disorders, pharmacological treatments for SUDs target the
full constellation of the disorder rather than individual symptoms. Further, phar-
macological treatments of SUDs are typically specific for the class of drug that
is abused (e.g., opioids, nicotine). Nevertheless, several studies have examined
the effects of medications specifically on ratings of craving for a particular drug.
For example, Courtney et al. (2016) reported that naltrexone (opioid antagonist)
blocked conditioned craving to heroin cues, and Green and Ray (2018) reported
that varenicline (nicotinic partial agonist) dampened cravings for tobacco ciga-
rettes. In other studies, oxytocin blocked cravings elicited by cues related to both
cannabis (McRae-Clark et al. 2013) and tobacco cigarettes (Miller et al. 2016).
One promising new pharmacological intervention for SUDs, as well as other
psychiatric disorders characterized by intrusive thoughts, is n-acetylcysteine
(Dean et al. 2011; McClure et al. 2014; Minarini et al. 2017). N-acetylcysteine
is a precursor to the antioxidant glutathione, and it acts as a modulator of gluta-
matergic, dopaminergic, neurotropic, and inflammatory pathways. Although this
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
324 H. de Wit and A. K. Bershad
medication has not been approved for treatment, Dean et al. (2011) review prom-
ising results with this drug in the treatment of cannabis use, cigarette smoking,
cocaine addiction, and several other psychiatric conditions in which intrusive
thoughts play a role. However, larger studies on the effects of n-acetylcysteine
for cocaine addiction or cannabis addiction have been disappointing (Mardikian
et al. 2007; Gray et al. 2017).
Cognitive
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Pharmacological Interventions 325
Neuropharmacological Considerations
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
326 H. de Wit and A. K. Bershad
drugs that block dopamine function are also the primary basis for pharmaco-
logical treatment. As noted above, however, the role of dopamine specifically
in intrusive thoughts, separate from other symptoms of psychotic disorders, is
not known.
Another consideration in the pharmacological treatment of intrusive
thoughts is the temporal characteristics of dosing. Many psychiatric medica-
tions, such as antidepressant and antipsychotic drugs, are prescribed for daily
use over extended periods of time. Recently there has been a growing interest
in medications that can be used with a single or small number of administra-
tions. Examples of these are the use of ketamine for depression and, more
controversially, the use of psychedelic drugs in the treatment of anxiety, de-
pression, or substance abuse. How exactly single, high doses of psychedelic or
other psychoactive drugs improve persistent psychiatric symptoms is an excit-
ing and challenging new direction of research. Do their effects depend on the
psychological experience, or are they related to neuropharmacological changes
in brain function? Finally, there is also an important role for drugs used in
combination with a behavioral intervention, such as the use of MDMA during
psychotherapy sessions or the use of beta-adrenergic blockers during presenta-
tion of drug-related memory cues. These different modes of administration of
psychiatric medications offer promising new approaches to combining phar-
macological with behavioral interventions.
Conclusions
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Pharmacological Interventions 327
antagonists are typically the first line of treatment for schizophrenia or sero-
tonin reuptake blockers for depression, yet physicians often sample from a
range of medications on an individualized basis to find effective treatments.
The range of pharmacological treatments is partly because patients vary in
their response to medications, as well as because the symptoms within a dis-
order vary across individuals and over time. Further, diagnoses and disorders
are in reality the composite of individual symptoms, and thus the individual
symptoms are rarely the target of medication development. It is thus difficult to
determine which medications are effective specifically for an individual symp-
tom, such as intrusive thoughts. Further, most psychiatric medications target
a constellation of symptoms involved in the disorder rather than any specific
symptom. For example, SSRIs are prescribed for depression to improve all
the components of the disorder, which may include mood, sleep, appetite, and
energy levels. This makes it difficult to examine published studies into the effi-
cacy of various medications in the treatment of intrusive thoughts. Prospective,
controlled studies with specific intrusive thought-related outcome measures
are necessary as a next step in investigating these questions.
We conclude with a comment about the recent introduction of a novel form
of pharmacological treatment for psychiatric disorders: the use of single, high
doses of “psychedelic” drugs. Drugs such as ketamine, MDMA, and psilocy-
bin, which were formerly only considered in the context of nonmedical use, are
now being tested in therapeutic settings. Remarkably, these drugs appear to be
effective after just one or two single administrations, in contrast to other medi-
cations that are used on a daily basis. It remains to be determined how single
doses of these drugs can produce lasting beneficial effects on, for example, ma-
jor depression, end-of-life anxiety, or PTSD. It also remains to be determined
whether, and how, these novel treatments affect intrusive thoughts in particular.
This is an exciting and promising new direction for psychiatric medications.
Acknowledgments
HdW was supported in part by NIDA DA02812 and AKB was supported by a training
grant from the National Institute of General Medical Sciences (2T32GM007281). We
thank Larry Price for helpful comments on the manuscript.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
16
Developing a Neuromodulation
Tool to Suppress
Intrusive Thinking
Things We (Think We) Know and
Things We Need to Figure Out
Abstract
Intrusive thinking is a core feature in multiple psychiatric diseases, including obsessive-
compulsive disorder (OCD), posttraumatic stress disorder (PTSD), substance use dis-
order (SUD), and Tourette syndrome. These diseases are not only bound by intrusive
thinking, they also share similar disruptions in the functional architecture of the brain,
including frontal-striatal-thalamic circuitry which is involved in salience attribution and
shifting attention. As more is learned about the neural circuit dysfunctions involved in
the initiation, maintenance, and attention to intrusive thoughts, it may become possible
to develop noninvasive neuromodulation approaches to attenuate the presence of these
thoughts or the morbidity associated with their existence in individuals. This chapter
focuses on transcranial magnetic stimulation (TMS) as a tool to induce causal change
in behavior, cortical excitability, and frontal-striatal activity. An overview is provided
of the cortical and subcortical areas that are often implicated in intrusive thinking, using
examples from Tourette syndrome, OCD, PTSD, and SUD. The hypotheses presented
can be generalized past TMS to other invasive and noninvasive forms of neuromodula-
tion. In conclusion, key questions are posed to move the field forward.
Introduction
As discussed in the other chapters of this book, there are many operational
definitions for intrusive thoughts. While it is difficult to unify these defini-
tions into one common framework, a familiar theme is present in all of them:
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
330 C. A. Hanlon and L. M. McTeague
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Developing a Neuromodulation Tool 331
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
332 C. A. Hanlon and L. M. McTeague
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Developing a Neuromodulation Tool 333
cortical excitability (via EEG and functional MRI), very little is known about
the effects of rTMS on neuropharmacology. The most cited studies in this do-
main have been done using positron emission tomography (PET) imaging,
wherein the radioligand is given to the participant and then the TMS stimula-
tion is delivered before the participant goes into the PET scanner. Using PET
imaging, Strafella et al. (2001) have demonstrated that dorsolateral prefrontal
cortex (dlPFC) stimulation leads to an increase in dopamine binding in the
caudate. They also showed that when 10 rTMS is delivered to the left primary
motor cortex, increases in dopamine are seen in the ipsilateral, left putamen.
Using magnetic resonance spectroscopy, several studies have demonstrated the
effects of TMS on cortical γ-aminobutyric acid (GABA) and glutamate (Stagg
et al. 2009; Vidal-Pineiro et al. 2015; Iwabuchi et al. 2017). One of the most
cited studies, by Stagg et al (2009), demonstrated that the attenuating effects
of inhibitory, continuous theta burst stimulation (cTBS) on cortical excitabil-
ity are related to an increase in GABA at the area of stimulation rather than
a change in glutamate. Recently, Iwabuchi et al. (2017) showed that a single
session of excitatory, intermittent theta burst stimulation (iTBS) to the dlPFC
leads to a decrease in the GABA/glutamate ratio in both the dlPFC and in the
insula, suggesting that it is possible to modulate paralimbic cortex through
superficial PFC stimulation.
The following principles need to be considered as TMS therapeutic strate-
gies are developed to address intrusive thinking.
With a growing number of TMS coil designs, the depth at which stimulation
should occur has become increasingly complex. The focality of TMS is related
to the shape of the coil. There is a substantial body of literature devoted to
computational modeling of electric field distributions associated with differ-
ent coil shapes. In one of the most comprehensive papers, Deng et al. (2013)
investigated the focality and penetration depth of 50 existing TMS coils.
Their computational models revealed that typical figure-of-eight coil designs
affected approximately 10 cm2 of cortical surface, circular coils affected ap-
proximately 50 cm2, and H-coil designs affected approximately 100 cm2. Most
flat figure-of-eight and circular coil designs had penetration depths of 1–2 cm2,
whereas the H-coil designs had consistently higher depths of 2–3 cm2. A single
TMS pulse from a standard figure-of-eight coil stimulates a 12.5 cm2 area,
which is approximately 1/125 (0.8%) of the cortical surface area. By compari-
son, deep brain stimulation can be at least an order of magnitude more precise
than the most focal TMS coils available, with stimulation volumes ranging
from 1–20 cm, depending on the electrode configuration (Wei and Grill 2005).
Electroconvulsive therapy, on the other hand, appears to effect 94% of the
brain and magnetic seizure therapy effects 21% of the brain (Lee et al. 2016).
To put the focality of TMS in context with something that is meaningful to the
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
334 C. A. Hanlon and L. M. McTeague
average curious member of the public, 1/125th of the cortical surface is roughly
analogous to the surface area of India (or half of Australia) relative to Earth
(Hanlon 2017).
R L
Figure 16.2 Polysynaptic effects: (a) Single pulses of TMS delivered to the hand
knob of the primary motor cortex are able to transmit information down the cortical
spinal tract, which crosses the synapse in the ventral horn, leading to contraction of
the efferent target muscle in the hand, measured with motor-evoked potentials. (b) This
polysynaptic engagement can be demonstrated in the cortex as well, wherein single
pulses of TMS delivered to the left dorsolateral prefrontal cortex (dlPFC) lead to el-
evated BOLD signal in the dorsal striatum and ventral cingulate, whereas (c) TMS to
the left frontal pole leads to BOLD signal in the ventral striatum, dorsal cingulate, and
anterior insula. Adapted after Hanlon (2017).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Developing a Neuromodulation Tool 335
As stated above, when single pulses of TMS are delivered in rapid succession
(rTMS), it is possible to change cortical excitability and various behavioral
phenomena for a relatively brief period of time (e.g., 30 minutes to several
hours; see Figure 16.3). These effects appear to be frequency dependent: low-
frequency, continuous stimulation decreases cortical excitability whereas high-
frequency, intermittent stimulation leads to an increase in cortical excitability
(reviewed in Fitzgerald et al. 2006; Thickbroom 2007).
One of the first studies in this field was conducted by Pascual-Leone et al.
(1994), who discovered that 20 pulses at 10 Hz and 20 Hz stimulation over the
motor cortex produced an increase in the amplitude of the MEP, suggesting this
frequency increases cortical excitability. Chen et al. (1997) then demonstrated
that 15 minutes of 0.9 Hz TMS stimulation (810 pulses) to the motor cortex
would decrease motor cortex excitability. In a sample of 14 individuals, 1 Hz
TMS to the motor cortex for 15 minutes decreased the MEP by 20% for at
least 15 minutes after stimulation. These data are compatible with preclinical
electrophysiology studies which have demonstrated that 1 Hz stimulation in-
duces long-term depression of neural activity in slice preparations of the motor
cortex, visual cortex, and hippocampus.
While 10 Hz and 1 Hz TMS are still widely used, a unique bursting fre-
quency known as human theta burst stimulation (TBS) has now gained sig-
nificant traction in the field. Human TBS was first evaluated by Huang et al.
(2005). Leveraging data from preclinical literature, which demonstrated that
10 s
10 Hz
5s
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
336 C. A. Hanlon and L. M. McTeague
electrical stimulation of cortical slices in 100 Hz bursts five times per sec-
ond (known as theta burst) can induce long-term plasticity (Bear and Malenka
1994; Malenka and Bear 2004), Huang et al. performed a clinical TMS study
wherein TMS pulses were delivered to the motor cortex in 50 Hz bursts five
times per second (human TBS). When TBS was delivered continuously for
600 total pulses, it decreased motor cortex excitability. When TBS was de-
livered in an intermittent pattern (2 sec on, 8 sec off) for 600 pulses, excit-
ability increased. The effect sizes of these brief continuous TBS (40 sec) and
intermittent TBS (190 sec) paradigms are comparable to studies of 1 Hz and
10 Hz. However, many publications have recently shown that there is high in-
terindividual variability in TBS response, which has led to some caution in the
reliance on this stimulation protocol (Vernet et al. 2014; Jannati et al. 2017).
A large body of literature demonstrates that the effects of TMS on behavior are
brain-state dependent and may be amplified by priming the brain with either a
behavioral task or brain stimulation (Opie and Cirillo 2017). One of the earli-
est studies in this field was by Iyer et al. (2003), who demonstrated that the
attenuation of cortical activity with 1 Hz TMS can be amplified by priming
the motor cortex with 6 Hz TMS. This was expanded to studies in the motor
system and visual system, which demonstrated that there were brain-state de-
pendent effects of TMS on cortical excitability (Silvanto et al. 2007, 2008a, b).
Additionally, priming the brain with continuous TBS may enhance efficacy of
intermittent TBS (Opie et al. 2017).
Although this body of research existed in sensory and motor control litera-
ture, it has only recently been harnessed by the clinical TMS research field.
Whereas the recent FDA approval of TMS for OCD requires a behavioral
prime, for example, neither the brain state nor the behavioral state of the indi-
vidual was accounted for during the initial multicenter clinical trials of TMS
for depression. This represents a latent opportunity for us to improve outcomes
and minimize some of the interindividual variability that is observed in pa-
tients receiving clinical TMS treatment.
Within the addiction literature, a large clinical trial demonstrated that ex-
posing a smoker to smoking cues (behavioral prime) before TMS amplified
the effects of TMS on smoking cessation (Dinur-Klein et al. 2014). In this pro-
spective, double-blind, sham-controlled study, 115 regular cigarette smokers
were randomized to receive ten daily treatments of TMS. Immediately before
each session, half of the participants were presented with visual smoking cues:
cigarette consumption and nicotine dependence were reduced, and the effects
were greatest in individuals that were exposed to smoking cues. In PTSD treat-
ment, priming a trauma memory at the outset of each rTMS session has also
been shown to enhance TMS effect sizes (Isserles et al. 2013). In this study,
thirty PTSD patients were randomized to one of three groups: sham rTMS,
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Developing a Neuromodulation Tool 337
One of the key advances in the neuroimaging literature over the last twenty
years is that brain regions organize their activity into coherent functional net-
works (Figure 16.4). Through functional magnetic resonance imaging (fMRI),
these networks appear as correlations of the low-frequency fluctuations in
BOLD signal between brain regions. Many networks were originally identified
via data-driven methods from brain activity at rest, and are called resting-state
networks. However, these networks reliably appear in ongoing brain activity
during tasks, and meta-analyses of task-based activation also reveal consistent
functional networks similar to those identified at rest.
Several functional networks have been studied extensively that have rele-
vance to intrusive thinking: the default mode network (DMN), containing the
medial PFC and posterior cingulate; the salience network (SN), containing
the anterior cingulate and anterior insula; and the executive control network
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
338 C. A. Hanlon and L. M. McTeague
pre-SMA/dACC
(OCD target)
dlPFC
(executive control
mPFC network node)
(default mode
network node)
Thalamus Striatum
Globus Pallidus Interna
Ventral
Figure 16.4 Candidate neural circuits amenable to modulation for intrusive thinking.
Typically, therapeutic neuromodulation approaches require that a specific neural system
be identified. The therapeutic strategy can then target the nodes of this neural system.
This can be done invasively (e.g., through deep brain stimulation) or noninvasively
(e.g., using TMS, transcranial electrical stimulation, pulsed ultrasound). Three corti-
cal nodes may be putative targets for neuromodulation based on their role in intrusive
thinking and their striatal-thalamic connectivity: the medial prefrontal cortex (mPFC,
red), the left and right dorsolateral prefrontal cortex (dlPFC, yellow), and the presupple-
mentary motor area (pre-SMA, blue). The striatal, pallidal, and thalamic nodes of these
circuits are shown in the lower panels. Adapted after Morris et al. (2016).
(ECN), containing the dlPFC and posterior parietal cortex. DMN is the best
known and most studied of these functional networks. It serves various in-
trospective functions related to intrusive thinking, including mind wander-
ing, recollection and prospection, rumination, and self-reflection. Other “task
positive” networks act in opposition to the DMN. These networks activate
during behaviorally regulated task performance and externally focused cog-
nition. For example, the SN activates for transitions from introspection to
task performance as well as during task initiation and switching. The ECN
is involved in cognitive control, working memory, and in tasks governed by
external stimuli, whereas functional connectivity in the DMN is typically
high during tasks of internal monitoring. In this manner, the ECN and DMN
are considered anticorrelated networks.
Etkin and colleagues demonstrated the central importance of the SN as a
common neural substrate across psychiatric illness categories (Goodkind et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Developing a Neuromodulation Tool 339
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
340 C. A. Hanlon and L. M. McTeague
as a target for PTSD, with mixed success, intrusive thinking is not often re-
ported as a primary outcome measure. Consequently, with mixed results from
the dlPFC, it is still unclear if targeting this area is likely to improve intrusive
thoughts.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Developing a Neuromodulation Tool 341
obsessional thoughts about dirt). It was assumed that this external cue places
the brain in a primed state, which would then be modulated by the TMS.
Based on these empirical results from rTMS experiments and from a strong
preclinical foundation regarding manipulation and reconsolidation of mem-
ories, it is likely that any neuromodulation approach for intrusive thinking
should involve putting the individual in a state where the intrusive thoughts
are present and perhaps bothersome.
Given that intrusive thoughts are frequently transient and share some of the
same temporal properties of a seizure (e.g., largely unpredictable onset time
but with some known triggers, an episodic disease feature rather than a stable
state as is seen in mood disorders or chronic pain), a closed-loop system may
be more effective and appropriate for intrusive thinking than an open-loop neu-
romodulation approach (Figure 16.5).
Open-loop neuromodulation typically refers to a device that provides a
fixed stimulation protocol over a fixed period of time. Currently, most invasive
and noninvasive neuromodulation approaches are open loop (e.g., TMS, DBS).
It is easy to see the value of a closed-loop system, which could include “sens-
ing technology” to detect changes in the brain state, and then dynamic stim-
ulation settings, which could adjust to the individual patient’s neural needs.
Closed-loop stimulation technology has shown promise as a treatment for vari-
ous diseases (Widge et al. 2017). The most successful closed-loop system in
clinical trials thus far has been developed, FDA approved, and is now deployed
for use in intractable epilepsy. The leads of this device (NeuroPace) are im-
planted into an epileptic focus in the brain. It monitors activity in the areas,
can detect the prodrome of seizure activity, and when a seizure begins it will
deliver stimulation to block the growth of that activity. This is referred to as re-
sponsive neurostimulation and received approval from the FDA in November,
2013. In the years since its approval, it has been successfully employed in
hospitals throughout the United States: several trials demonstrated a 53% sei-
zure reduction after two years and a 70% median seizure reduction after five
years (Geller et al. 2017). Closed-loop stimulation has also been used in the
spinal cord for pain management (the RestoreSensor™ System, Medtronic,
Minneapolis, MN). Although no noninvasive, closed-loop brain stimulation
devices have been approved for clinical use yet, there is extensive growth in
this field (Bergmann et al. 2016).
There are also non-device-based closed-loop neuromodulation strategies.
Simpler versions include a “human in the loop”: clinicians observe the re-
corded brain signals and provide manual adjustment of the stimulation rather
than the device automatically self-adjusting. Other techniques such as real-time
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
342
Figure 16.5 Open- and closed-loop neuromodulation options for intrusive thinking. Given the temporal profile of intrusive thinking, a closed-
loop neuromodulation approach may be more beneficial—one that includes the ability to detect aberrant brain states (or behavioral states) and
adapt its stimulation properties accordingly (e.g., turning itself on/off, modifying the frequency, changing the amplitude). A critical element in-
volved in closed-loop technologies, however, is the reliability of the underlying biomarker it is trying to sense. In practice, standard “open-loop”
techniques such as TMS and DBS are actually closed-loop systems, because the patient always returns to the clinician to have the parameters ad-
justed. New generation closed-loop techniques, however, may enable these adjustments to be made in real time, responding to endogenous neural
changes and increasing the “agency” of patients by enabling them to modify the settings based on their own perceptions. This may be particularly
useful for intrusive thinking.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Developing a Neuromodulation Tool 343
fMRI feedback, EEG feedback, and mindfulness strategies may also be useful
approaches for intrusive thinking.
4. What is the best treatment strategy: preventing the initiation of intrusive
thoughts or increasing the ability of the patient to shift their attention away
from the thoughts?
When designing a neuromodulation approach for intrusive thoughts, the
neural regions targeted and the dosing parameters used will likely be differ-
ent if the goal is to help individuals shift their attention away from intrusive
thoughts than if the goal is to stop them from happening. If the goal is to
stop the initiation of intrusive thoughts, it is possible that targeting the DMN
(perhaps using TMS directed at the medial PFC) might be a fruitful strategy.
Alternately, if the goal is to enable individuals to shift their attention away
from thoughts which target the salience network, a deep form of TMS di-
rected at the cingulate or insula might be the best strategy, given its role in
set-shifting and attributing value.
Conclusion
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
344 C. A. Hanlon and L. M. McTeague
Acknowledgments
Figure 16.4, from Morris et al. (2016), is used under the terms of the Creative Commons
Attribution 4.0 International License (CC-BY).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
17
Abstract
Group photos (top left to bottom right) Judson Brewer, Harriet de Wit, Aurelio
Cortese, Martin Paulus, Emily Holmes, Judson Brewer, Damiaan Denys, Colleen
Hanlon, Peter Tse, Jens Schwarzbach, Martin Paulus, Colleen Hanlon, Aurelio Cortese,
Harriet de Wit, Judson Brewer, Peter Tse, Damiaan Denys, Martin Paulus, Harriet de
Wit, Jens Schwarzbach, Emily Holmes
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
348 J. Brewer et al.
Introduction
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Interventions and Implications 349
Many types of mental events can intrude into consciousness. Verbal and non-
verbal thoughts, mental images, impulses, memories, emotions, desires, and
dreams can all reset the contents of consciousness and be experienced as
unwanted or intrusive. Do all such events have the potential to be clinically
relevant? How does the minimum definition, which primarily encompasses
psychiatric symptoms, relate to the maximum definition, which relates to ev-
eryday experienced phenomena? To what extent do they overlap or deviate
from each other? To what extent are they qualitatively, or perhaps only quanti-
tatively, different from each other? Importantly, if clinically relevant intrusive
thinking takes different forms or domains (e.g., verbal vs. imagery), what are
the implications for treatment? Different treatment approaches may be needed
or optimized for different domains.
In our discussions, we juxtaposed these two definitions next to each other—
one with a psychiatric clinical purpose, the other with a psychological fun-
damental purpose—but wish to emphasize that intermediate viewpoints are
possible. To address our group’s topic, however, we found these artificially
contrasting definitions helpful. A discussion of the philosophical and social
implications of defining the phenomenon is included later in the chapter.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
350 J. Brewer et al.
defining features and may causally drive other symptoms (see Holmes et al. as
well as Hanlon, this volume). In others, intrusive events may not be the defin-
ing feature of a disorder but rather one of many criteria used to reach a diagno-
sis. For example, craving is a symptom of substance use disorder (SUD), but
SUDs can and do appear without cravings. Similarly, suicidal ideation appears
in major depressive disorder, but the disorder can occur without it. By focusing
on a single symptom, it may be possible to relate the clinical manifestation of
an intrusive event to a dysfunction of neural circuits controlling normal brain
function.
Given the heterogeneous nature of intrusive events associated with different
psychiatric disorders, we recommend that a research program be developed
to examine whether subcategories of intrusive events exist; this information
is needed to establish an association with the underlying neurobiology. More
importantly, intrusive events with a different neurobiological signature may
require distinct treatment approaches. Given the different dimensions of an
intrusive event (e.g., prior experience and expectancies, precipitating events,
temporal sequence of events, emotional breadth, contextual features, psycho-
logical consequences), we offer two clinical vignettes to illustrate the clinical
manifestations of intrusive events in actual patients and demonstrate the inher-
ent complexities.
From an index traumatic event, patients with PTSD typically experience two
or three different, highly vivid intrusive memories both in visual and other
sensory modalities (Grey and Holmes 2008). For example, after a traumatic
road accident, a person might experience vivid intrusive visual mental images
of an oncoming red truck, which originated from the moment just before the
accident. This intrusive image is highly distressing and often associated with
the strong emotions that occurred at the time of the trauma (fear, helpless-
ness). Additional multimodal sensory images may originate from the same
event. For example, the visual memory is also multimodal, comprising sight
of the person’s hand on the driving wheel accompanied by the sound of glass
breaking and the smell of burning. The memory may also be associated with
secondary emotions such as horror and guilt. Well after the event has oc-
curred, an otherwise innocuous visual stimulus, such as a red front door, may
remind the person of the red truck, thus triggering an emotional response.
Emotional states may also serve as triggers for the traumatic event: when a
person experiences a feeling of helplessness about an unrelated event, this
may elicit the feelings of helplessness associated with the traumatic memory.
These experiences may be especially disturbing because the patient has not
associated these external or internal cues to the occurrence of the intrusive
memory, so that they occur as if without warning and as both unpredictable
and uncontrollable. Thus, even though these intrusive memory experiences
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Interventions and Implications 351
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
352 J. Brewer et al.
Obsessive-Compulsive Disorder
In OCD, intrusiveness coincides with the feeling of “being out of control” and
is experienced in different phenomenological domains, as OCD develops over
time. OCD is a process with different clinical stages rather than one single
stage. These stages follow a dialectic interaction in which an intrusive event
elicits a response and the response amplifies the intrusive event. Through dif-
ferent neurobiological adaptations, the course of OCD eventually worsens.
Thus, OCD should be regarded as a disease process that develops through the
amplifying interaction between (the reflection and resistance of) the person
(mind) and the disorder (brain).
Take, for instance, a young mother who recently gave birth to her first child.
She carries an enormous burden of being solely responsible for a helpless and
vulnerable life. Her husband leaves daily for work, leaving her alone at home
with the child. Seeing her young baby in the crib, a thought appears in her
mind: she imagines that she could strangle her baby in the crib and that because
she is alone, no one could prevent her from acting on that thought. In summary:
1. The mere presence of this thought is intrusive because it occurs against
her will. She feels out of control and is unable to volitionally control
her thinking.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Interventions and Implications 353
2. She is worried because the idea of strangling her baby does not
match the ideal of motherhood she maintains and strives to achieve.
The content of the thought is intrusive because it is not in line with
her identity or expectancy. It is ego-dystonic in that it does not
match her self-image, thereby giving rise to mental discomfort and
distress.
3. She wonders whether she really could strangle her baby. How can she
be certain, given the fact that humans are notoriously unpredictable,
that she won’t in fact destroy that which she most treasures? It seems
that the freedom to commit such a terrible act is in itself so disturbing
that it causes the thought repeatedly to reoccur. In addition, moral im-
plications of the initial thought intrude as well.
4. She feels anxious because the thought confronts her with feelings of
being out of control. The emotional value (anxiety) of the thought is
intrusive. The presence, content, implication, and emotional value of
the thought all have an intrusive quality. Although she actively resists
the thought because it annoys her and feels intrusive, the very process
of reflecting and/or resisting the thought reinforces the frequency and
strength of the intrusive thought. Her efforts to eliminate the thought
may, in fact, enhance its occurrence.
5. The thought becomes obsessional, and her attention is completely
drawn to that one single thought. Obsessionality is a dysfunction of
intentionality: the incapacity to shift focus or attention to another topic,
due to a stronger and longer intentional relation with the mental act.
The thought is intrusive because of its obsessive nature.
6. She cannot suppress the thought; moreover, she is compelled to think
about her obsession. Compulsivity is a dysfunction of sense of agency:
she is forced to think about the intrusion, contrary to her willpower.
The thought is intrusive because of its compulsive nature.
7. Gradually the thought becomes more present and repetitive; it loses its
original meaning, but remains an intrusion because of its duration and
repetition. The thought is intrusive because of its new form or appear-
ance; it is now a full-blown obsession.
8. Obsessions are answered with compulsions. (Note: both obsessions
and compulsions are intrusive, with both having obsessional and com-
pulsive qualities.) Though initially successful in reducing anxiety,
these compulsions gradually become intrusive since the acts have to be
performed compulsively.
9. Eventually, the anticipatory power of the intrusion becomes so over-
whelming that reality testing is disturbed. She does not know anymore
whether she has or has not strangled her baby. Thoughts may become
delusion-like, with psychotic features.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
354 J. Brewer et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Interventions and Implications 355
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
356 J. Brewer et al.
experience in daily life (James et al. 2016c; Lau-Zhu et al. 2019). Recording
modes can range from pen and paper, to an SMS or an online interface. Diaries
have the merits of being able to be precisely tailored to the research question in
mind; they are sampled in real time and are thus less prone to memory biases
inherent in retrospective self-report. One could imagine that apps will become
useful in this regard.
Although intrusive events may be a component of worry, simultane-
ous administration of worry and intrusive event questionnaires yielded dif-
ferent factor structures. Nevertheless, the factor structure for the strategies
used to counter the thoughts were highly similar for both types of thought.
Furthermore, regression analysis identified interesting relationships between
the strategies, the thought characteristics, and appraisal of intrusive thoughts
and worry (Langlois et al. 2000a, b).
It appears that concern about the personal meaning of the thought is a unique
dimension for obsessive intrusive thoughts (Clark and Claybourn 1997). There
is, however, fundamental disagreement as to whether clinical and nonclinical
intrusive events can be conceptualized along a continuum. Some researchers
focus on thought content: Belloch et al. (2004) found that the ten most fre-
quently occurring thoughts were related to accident, harm, sex, and aggression.
Others focus on the process characteristics, thus emphasizing the intrusive
aspect of the thought (Rachman and de Silva 1978). Interestingly, Lee et al.
(2005) found that the most upsetting intrusive thought is often autogenous; that
is, such intrusions come abruptly into consciousness without identifiable evok-
ing stimuli and are perceived as ego-dystonic, aversive enough to be repelled,
and include sexual, aggressive, and immoral thoughts or impulses.
Intrusive thoughts are thought to be closely related to dysfunctional be-
liefs (Obsessive Compulsive Cognitions Working Group 2003). Thus, there is
an urgent need to assess underlying belief systems associated with intrusive
events. These include:
• Over-importance of thought: beliefs that the mere occurrence of an in-
trusive thought marks its significance.
• Need to control thoughts: beliefs that one can and should exercise com-
plete control over unwanted intrusive thoughts, images, and impulses.
• Perfectionism: beliefs that a perfect response or solution to every prob-
lem is necessary and that even a minor mistake can lead to serious
consequences.
• Inflated responsibility: beliefs that one is liable for causing and/or pre-
venting significant negative outcomes for self or others.
• Overestimated threat: beliefs involving exaggerated estimates of the
probability and/or severity of harm to self or others.
• Intolerance of uncertainty: beliefs that it is necessary to be certain and
that unpredictability and ambiguity should be minimized as much as
possible.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Interventions and Implications 357
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
358 J. Brewer et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Interventions and Implications 359
Treatment Approaches
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
360 J. Brewer et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Interventions and Implications 361
As discussed, very few treatments exist that explicitly target intrusive events
as an aspect of a disease, and for those that do exist, it is not clear how they
actually modify intrusive events. Although intrusive events are included in the
DSM-5 criteria for many disorders, they have not yet been targeted as a distinct
treatment domain.
Treatment innovation is essential and may benefit from being mechanisti-
cally driven and by combining treatment modalities, such as combining phar-
macology with a psychological/behavioral approach (Holmes et al. 2018).
Before going into detail about specific treatment modalities, we discuss how
interventions that target intrusive events might be used to modify core aspects
of an intrusive event: gating, error correction, salience, and evaluation.
Gating Errors
Within psychiatric diseases, intrusive events may result from too much infor-
mation being allowed into awareness, either in a global sense or by a selective
memory gaining access to awareness through a selective gate. This broad con-
struct could be applied to many psychiatric diseases. Intrusive verbal thoughts
associated with general anxiety disorder (see earlier discussion) may be due to
a “weak” gate that allows a lot of information in and ultimately leads to stress
and anxiety. Intrusive image-based thoughts associated with PTSD may be due
to a faulty assignment of salience and/or evaluation that emerges for the trau-
matic event, which allows them to enter into or persist in awareness more eas-
ily than nontrauma-related thoughts. Intrusive thoughts of drug cues in SUD
may be construed as a combination of faulty gating and salience, compounded
by reinforcement learning.
Error Signal/Detection
Intrusive events may also result from a heightened error signal during oth-
erwise normal thought. To achieve ordinary activities of daily living, mental
processes must stay on course and not be derailed by irrelevant streams of in-
formation that constantly come into our brain. A lot of information is processed
on a subconscious level, wherein only the stimuli that are most different from
our expectations are processed consciously. For example, when walking down
the street, an individual is able to maintain posture and balance, typically via
subconscious processing that has become highly automatized. Only when an
event occurs that violates expectations (e.g., when we stumble or suddenly see
an unexpected object in our path) is the walking process brought to the level of
consciousness via bottom-up mechanisms. Alternatively, our own cascade of
ongoing neural processing related to other aspects of life, which are not directly
associated with the goal, can also interrupt automatic behaviors, as when we
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
362 J. Brewer et al.
suddenly remember that we were supposed to pick up the kids at school. Thus,
bottom-up and top-down internal interrupts exist as well as external interrupts.
The difference between our expectations of internal and external stimuli and
the actual perception of these stimuli can be considered a prediction error.
It is possible, for example, that verbal or visual intrusions in PTSD may
be related to a momentary yet large prediction error. That is, at times, the per-
ceived difference between the thoughts associated with the trauma and the
typical subconscious narrative that is ongoing in an individual’s mind is large
(high prediction error) and triggers the individual to switch their attention to
those momentary thoughts. This is a particularly attractive hypothesis given
the sparse temporal nature of intrusive events in these patients. In daily living,
for instance, many individuals with PTSD are able to function relatively well
in between intrusive events of their trauma. They are often able to work and
care for families, and to conduct their lives for some days without having an
intrusive event. The frequency of occurrence of an intrusive memory of trauma
for some patients may be infrequent (e.g., once a fortnight) whereas for oth-
ers with more severe levels of PTSD, it may be up to every hour. Typically,
measurement tools of PTSD capture this rate of occurrence, although more
research is needed.
Here we address emerging ideas for future treatment innovation based on
the model proposed in the Appendix. Assuming that these reward prediction
errors are coded in a specific neural network, a technology capable of sensing
the magnitude of the violation between expectations and actual input may be
able to change activity in this network in a dynamic manner, pushing it back
into the intended state. This type of closed-loop neuromodulation is currently
used in the treatment of epilepsy. Briefly, sensing and stimulating electrodes
are placed in the brain of individuals with intractable epilepsy in the vicinity
of the seizure focus. The device has a certain “tolerance” for background vari-
ability in the neural activity in the vicinity of the electrode. Once a critical level
of variance is detected (high variance), the device is able to stimulate the brain
and push it back into a healthy state (low variance), thus avoiding the cascade
of a seizure (for further discussion, see Hanlon and McTeague, this volume). It
is easy to see how a similar approach could be used to abolish intrusive events
in a dynamic manner in individuals with PTSD. In veterans with PTSD, for ex-
ample, a device could be trained to identify the “background” levels of activity
present in normal life as well as activity associated with intrusive war-related
memories, and permit the device to push the system back into the healthy range
of error. One problem with this approach, however, is that large prediction
errors are a crucial element for flexible human behavior. So, an autonomous
closed-loop system would have to be able to detect differences specific to the
trauma memory. While this seems like a tall order, because these intrusive
events are so debilitating to the patients, it is reasonable to imagine that the am-
plitude of their prediction error is so large that the device would be able to have
a very high tolerance threshold, and thus only produce stimulation during the
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Interventions and Implications 363
Evaluation
Pharmacological treatments may change the sensitivity of the gate that al-
lows items to enter into conscious awareness. For example, benzodiazepines
may blunt emotional response to negative thoughts in a nonselective manner.
Alternatively, there may be potential for combination treatments that couple
a behavioral approach (e.g., one that evokes intrusive events and thoughts,
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
364 J. Brewer et al.
such as cue exposure), to open the gate, with a pharmacologic agent, to blunt
the response.
Imagine if one could simply “zap” the brain and magically remove the in-
trusive event causing the distress. Such an idyllic scenario is (unfortunately)
still far away, but promising research is moving us closer. A whole new field of
treatment research has emerged around specific brain stimulation techniques
due to their noninvasive nature and simple theoretical motivation. The central
idea behind brain stimulation and modulation techniques takes a systems per-
spective, which considers the brain as the source of any behavior (the outcome
of brain processes). Behavior is therefore the dependent variable whereas brain
circuits, activity, or neurons are the independent variables. If we are to change
a pathological behavior, this systems-level interpretation implies that we have
to find, target, and modulate the brain process or instantiation that is generating
this specific behavior.
Two main approaches have produced important results in recent years:
TMS and neuroimaging-based neurofeedback. TMS is a technique that gen-
erates weak electrical currents within the brain through direct application of
electromagnetic flux over the scalp, noninvasively into brain tissue (see also
Hanlon and McTeague, this volume). At the electrophysiological level, TMS
induces changes in neuronal excitability, which can cause changes in behavior.
TMS has been recently approved by the FDA for use in the treatment of sev-
eral psychiatric disorders, such as addiction and OCD. Future studies need to
evaluate how TMS in combination with other therapeutic forms might be used
specifically to resolve intrusive events.
In neuroimaging-based neurofeedback, “neurofeedback” is defined as a
closed-loop procedure whereby online feedback of ongoing neural activity is
given to the participant for the purpose of self-regulation. Noninvasive neuro-
feedback can be implemented with several neuroimaging modalities. Some of
the most promising proof-of-concept cases derive from functional magnetic
resonance imaging (fMRI) and electroencephalography (EEG)-based neuro-
feedback (Sitaram et al. 2016; Taschereau-Dumouchel et al. 2018a; Keynan et
al. 2019). We will first consider fMRI, due to its potential as an acute interven-
tion (used once or a few times).
With fMRI, one simple option is to use the overall signal strength in one
predefined brain region. This may work well when the target is general (e.g.,
motor action initiation vs. no motor activity); in other cases, however, one
needs to access a specific representation, which by definition would not be re-
trieved through the overall activation level. Therefore, rather than focusing on
the overall activity level of a region (unspecific), machine-learning algorithms
now allow us to infer the precise activity pattern that corresponds to a unique
stimulus, object, or category (specific). This approach, borrowed from machine
learning, is termed multivoxel pattern analysis (MVPA) (Kamitani and Tong
2005; Norman et al. 2006). Going even further, methods have now reached
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Interventions and Implications 365
the point where we can infer the pattern of brain activity in a target participant
from brain activity patterns in surrogate participants (Haxby et al. 2011).
Although these advances have often remained outside the realm of clini-
cal applications, recent innovative efforts have been directed toward utiliz-
ing MVPA in the field of neuropsychiatric disorders. While mainly centered
around the development of markers for certain disorders or their subtypes, with
some remarkable results (e.g., Yahata et al. 2016; Etkin et al. 2019), some stud-
ies have explicitly targeted the prediction of intrusive events from fMRI activ-
ity patterns (see Holmes et al., this volume; Clark et al. 2014b, 2016).
Bringing MVPA to a real-time setting leads us to the concept of neurofeed-
back, which essentially monitors brain activity patterns over time while using
the machine-learning prediction as neuromodulatory input (e.g., to provide re-
wards). If an algorithm is able to detect intrusive events, the system may then
provide a reward to remodel the association between brain state and appraisal
or some other form of modulation to disrupt the cascade of neural events fol-
lowing the emergence of the intrusive event.
In terms of clinical development, due to the intrinsic nature of MVPA-based
designs, neurofeedback interventions can be easily used in a double-blind,
placebo-controlled way. Since we can find activation patterns that correspond
to specific, distinct mental representations, we can also let the algorithm ran-
domly choose which representation will be the target and which the control.
Neither the experimenter nor the patient has to be aware of the category for the
software procedure to be deployed effectively.
Research has shown that neurofeedback may work via reward processing,
learning, and control networks (Sitaram et al. 2016), and that it depends on
reinforcement learning processes (Shibata et al. 2018). During a typical ex-
periment, the machine-learning algorithm monitors activity in a selected brain
region and, at predefined time intervals, computes a (monetary) score reflect-
ing the likelihood that the current activity pattern resembles a template, target
mental representation. Over time, the brain learns to associate the occurrence
of such representation with rewards.
In two proof-of-concept studies, this technique was used to reduce fear
responses in healthy individuals conditioned toward simple visual stimuli
(Koizumi et al. 2016) as well as in participants with subclinical phobia toward
their feared object (Taschereau-Dumouchel et al. 2018a). The physiological
fear responses (amygdala reactivity and skin conductance) were diminished
only for the targeted representation, while a control stimulus elicited unchanged
fear responses. More recently, the same group reported on the feasibility of us-
ing decoded neurofeedback as a target treatment in PTSD (Chiba et al. 2019).
Importantly, throughout these studies, participants were not told about the link
between their brain activity and the amount of reward they received on a trial-
by-trial basis. When asked to make a forced choice about the target and control
categories, participants answered randomly (Shibata et al. 2018).
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
366 J. Brewer et al.
The aspects introduced so far raise one crucial point: the entire neurofeed-
back intervention can be applied without ever mentioning or displaying the
upsetting content or event to the patient. The enormous advantage that this
could bring is evident: reducing implicit, physiological fear responses could
create the foundation for subsequent, and more effective, behavioral or cog-
nitive therapies. Neurofeedback could then, as a second step, be applied for
maximal effects and learning.
Alternatively, neurofeedback modulation may provide the basis for damp-
ening a physiological, autonomic reaction, which may then facilitate subse-
quent cognitive or behavioral therapies. Current work on fMRI, MVPA-based
neurofeedback can be expanded to include the notion of dynamic brain state.
Indeed, utilizing a very simple correlation of activity fluctuation between two
brain areas, or a pattern of activity representing an object or category, may
be too simplistic, in particular if we are to target states that are more global
(e.g., attention, emotions, arousal, interoception). Spatiotemporal oscillatory
patterns describe relatively well these global states, distinguishing between
rest and task-based efforts (Vidaurre et al. 2017; Bolton et al. 2018).
Rather than acting on the intrusive events themselves, dynamics-based
neurofeedback interventions could target and modulate an overall affective or
cognitive state. Here, the rationale is different: prepare the brain to receive the
treatments targeting specific aspects of the pathology in the individual.
fMRI neurofeedback carries high costs, immobility, and the requirement for
specialized operating personnel. This greatly hampers the scalability of this ap-
proach, particularly if we think about going beyond acute treatments.
As such, EEG-based neurofeedback holds potential from a different per-
spective. New generation EEG headsets have relatively low costs, are small
in size, and are almost “plug-and-play” ready for use. We can envisage EEG
products in the near future that could be used autonomously by patients on a
daily basis and essentially without many constraints in terms of location or
function, allowing for real scalability. Proof-of-concept EEG neurofeedback
has been demonstrated with specific mental states while targeting deep brain
structures, such as the posterior cingulate cortex which is part of the default
mode network (van Lutterveld et al. 2017), training stress resilience through
electrical fingerprint (Keynan et al. 2019), as well as reduction, consolidation,
and personalization of lead placement (Pal et al. 2019).
A different form of feedback might therefore involve generating an exter-
nal interrupt when an undesirable thought pattern has emerged. For example,
MVPA of EEG signals might allow ruminative thoughts associated with de-
pression to be decoded. If detected, an EEG headset could then be designed
that might vibrate or emit a tone whenever rumination had exceeded, say, ten
seconds, bringing the patient back to the present.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Interventions and Implications 367
Combination Treatments
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
368 J. Brewer et al.
Although the original definition by Clark (2005) and its minimum and maxi-
mum derivatives are intuitively very recognizable, the appearance of clarity
is misleading. The problems that come to light reflect some important philo-
sophical and social assumptions.
What exactly is meant by unwanted? Is it a mental event that is unwanted
by the individuals themselves? Or does this represent what is unwanted by
members of society, relative to what counts as unacceptable, abnormal, or even
moral or immoral (e.g., expressions of sexuality or aggression)? Is what is un-
wanted not wanted relative to short-term individual goals, such as wanting to
drink water, eat a marshmallow, or court a colleague? Or is what is unwanted
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Interventions and Implications 369
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
370 J. Brewer et al.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Interventions and Implications 371
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
372 J. Brewer et al.
engagement with SMA may affect both brain structure and function, and in
turn make individuals more susceptible to the experience of both physiologi-
cal and pathological intrusive events. The investigation of mobile technology
on cognition is still in its infancy (Wilmer et al. 2017), and our knowledge
about the impact of this technology on intrusive events is nonexistent. Thus, an
important goal for future studies should be to determine whether such devices
could contribute to the exacerbation of psychiatric disorders, characterized by
frequent and/or severe intrusive events.
Another perspective on intrusive events and their pathology is the notion
that an individual who experiences a low level of need for control might not
experience intrusive events as problematic or in need of treatment. This raises
the question whether the fact that we are focusing on neuroscience, clinical
consequences, and interventions associated with intrusive events is a by-prod-
uct of our society’s focus on “over control.” Interestingly, individuals with
strong beliefs about controlling thoughts are more likely to experience dis-
tressing intrusions, both with and without meta-awareness, compared to people
with weaker beliefs (Takarangi et al. 2017). Thus, it would be interesting to
examine the frequency, severity, and clinical consequence of intrusive events
in different societies that place different emphases on cognitive control. Along
similar lines, an intrusive event might only be assessed as an unwanted pertur-
bation if the individual has a concept of causes and effects, which is a historical
consequence of the Enlightenment period of the seventeenth and eighteenth
centuries. In prior societies there was a greater emphasis on a teleological
framework within which experiences and events were interpreted through the
lens of their potential function, end, purpose, or goal. In this context, an in-
trusive event may be experienced as something that is necessary to lead to a
particular goal rather than an unwanted distraction. Thus, it may be interesting
to conduct a historical literature analysis that focuses on the characterization of
intrusive events before and after the Enlightenment period.
Let us return now to intrusive events and tie them in with the notion of
frameworks of meaning. If efforts to suppress intrusive events tend to exac-
erbate intrusive events, then perhaps efforts should be made concerning how
best to suppress such suppression, or, on the contrary, how best to facilitate
the expression of intrusive events so that the salient issue so expressed can
be processed in a healthy manner. Other cultures have created modes for the
expression of “forbidden” emotions. Ancient Greek drama often centered on
creating tensions that would then lead to emotional catharsis. Rather than sup-
press unwanted emotions, such as lust for forbidden objects of desire, as in the
play Oedipus Rex, these emotions were vented in a manner that was safe for
society. Even modern European cultures have aspects that are reminiscent of
catharsis. For example, Carnival is a venue for the expression of carnal desires
and behaviors that in other times would be regarded as deviant or dangerous.
How might catharsis be exploited in the context of existing or new therapeu-
tic methods? One possibility would be role-playing, where a traumatic event
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Interventions and Implications 373
Conclusion
Intrusive events are emerging as an area of interest that can help further our
collective understanding of basic brain function, psychiatric conditions, and
their treatment. Characterizing domains in which intrusive events manifest in
psychiatric disorders may help their characterization, diagnosis, and treatment.
Basic heuristic models can inform how normal brain function can go awry
due to faulty systems, including gating, salience, evaluation, and prediction
error detection. From these models, current and evolving treatments (and po-
tential combinations) can be tested for target engagement, specificity of effect
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
374 J. Brewer et al.
Our model begins with a sequence of goals that are represented in the brain
(Figure 17.A1). A mental workspace keeps the present goal in mind, allows
operations to take place over representations held in the workspace (area 4,
working memory), and takes into account prior knowledge, context, and other
system constraints. Salience helps the organism determine how important
a potential interrupt is and interacts with an evaluative system to determine
whether to stay on task or to interrupt it and, if so, which task to then prioritize
as the task to do next. A goal maintenance system helps the mental workspace
maintain the present goal using feedback loops that afford the minimization
of prediction error signals. This may happen via the enhancement of gating of
potential interrupt inputs, or the inhibition of the salience of potential interrupt
signals. The goal maintenance system evaluates deviations from the trajectory
leading toward fulfillment of the goal. Such prediction error signals are used
dynamically and cybernetically to correct the present trajectory to minimize
that error, similar to when a heat-seeking missile alters its path to hit its target.
External interrupts (not internally generated) may orient the organism to
unexpected inputs from the external world (e.g., when we hear a loud sound
or see a sudden motion): if the magnitude of the prediction error (i.e., between
what was expected and what in fact occurs) is large enough, the organism ori-
ents to the external stimulus.
Internal interrupts can be both bottom up and top down. Bottom-up sys-
tems that can generate interrupts include systems that maintain physiological
(e.g., hydration) and nonphysiological (e.g., happiness) goals that are separate
from the current goal. When salient enough, these interrupts provide inputs to
the mental workspace to force a reprioritization of what to do next (stay on
task or switch tasks). Subjectively these signals are experienced as, for ex-
ample, thirst, hunger, lust, or a need for oxygen, salt, or sleep. Other bottom-
up systems include reward/punishment and other evaluative systems but may
not have intrinsic homeostatic functions. These may be experienced as, for
example, fear or other emotions, such as anger. In addition, the memory sys-
tems may automatically retrieve memories which can then appear to “pop” into
consciousness.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Top down 2
internal interrupts: 4
6 Goal 1 4
Default mode intrusive thoughts 4 2
sequencing and
current goal stack 6
network 2
1
2 4
Working memory: 5
Exogenous
attentional Processing present
goal, the perceptual Current goal 5 6
interupts maintenance 5
world and world model,
executing current and switching 6 5 3
systems 5
mental operations 3 3
Lateral 3
External interrupts:
Intrusive disruption
Figure 17.A1 (1) Neural substrates: frontopolar cortex (e.g., Brodmann area 10). (2) Ventral attentional network: ventrolateral prefrontal cortex,
temporoparietal junction. (3) Numerous subcortical systems including amygdala, hippocampus, nucleus accumbens and ventral tegmental area,
brain stem, and thalamus. (4) Dorsal attentional network: dorsolateral prefrontal cortex, posterior parietal cortex, superior parietal lobule, and
intraparietal sulcus. (5) Cingulo-opercular control network: dorsal anterior cingulate, frontal operculum, basal ganglia; for emotional regulation
ventromedial prefrontal cortex (BA 12) and orbitofrontal cortex (BA 11). (6) Default mode network: anterior medial prefrontal cortex, posterior
375
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
376 J. Brewer et al.
Failure Modes
Normal Pathological
Gating Appropriate Too weak, too strong
Salience Appropriate Too little, too much, etc.
Evaluation Accurate (In)accurate
Prediction error Accurate (In)accurate, failure to reset
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
Interventions and Implications 377
enough when you think you didn’t), the thought can become pervasive, lead to
excessive worry, or inappropriately change the task to studying.
Salience signals can be faulty when studying seems much more appropri-
ate than necessary, and the task is changed from going to a movie to studying.
Another example of inappropriate salience occurs when one is studying, and
studying appears so much more important than sleep that one continues study-
ing until the moment of the exam, paradoxically hampering performance.
An inaccurate prediction error signal arises when one is on task, but none-
theless gets a signal that one is off task. For example, when someone is study-
ing for a test, one feels anxiety despite engaging in studying, which can para-
doxically undermine test preparation through perseverative worry (instead of
studying).
An additional pathway that can compound mode failures emerges from re-
inforcement learning to combine with the faulty elements described above. If
the threshold of the gate is too low, the interrupt changes the ongoing plan. The
change itself (because it is new) can be reinforcing. This reinforcement leads
to increased salience of the event that produced the change in plan. As salience
iteratively increases, the likelihood increases that the interrupt that is linked
to the new plan is going to disrupt other plans in the future (positive feedback
loop). If the new plan is now in place, the stronger salience of the interrupt sup-
ports its maintenance. For example, with perseverative worry, the worry led to
a change in plan (didn’t go to the movie), which then led the person to study
more and indeed feel better, which then functions as a reward signal that leads
to a reinforcement of perseverative worry in the future. Such a simple mecha-
nism could account for the common finding that OCD worsens with time. An
important question for future research into effective interventions will be how
to rein in this positive feedback loop afforded by reinforcement learning.
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1
From “Intrusive Thinking: From Molecules to Free Will,” edited by Peter W. Kalivas and Martin P. Paulus.
Strüngmann Forum Reports, vol. 30, Julia R. Lupp, series editor. Cambridge, MA: MIT Press. ISBN 978-0-262-54237-1